Fixed GPT-6 Sol and Luna, and the system prompt role of newer OpenAI models (#1010)

app/MindWork AI Studio/Models/ModelProfile.cs

@@ -92,6 +92,15 @@ public readonly record struct ModelProfile
     /// </summary>
     public ImageLimits Images { get; init; }
 
+    /// <summary>
+    /// Which role the system prompt is sent in.
+    /// </summary>
+    /// <remarks>
+    /// Stated by the rules of OpenAI's own models alone, and unknown for everything else: the other
+    /// providers have one way to send a system prompt, and the provider knows it without asking.
+    /// </remarks>
+    public SystemPromptRole SystemPromptRole { get; init; }
+
     /// <summary>
     /// Whether the model has every one of the given capabilities.
     /// </summary>

app/MindWork AI Studio/Models/ModelProfileChange.cs

@@ -55,6 +55,11 @@ public sealed record ModelProfileChange
     /// </summary>
     public ImageLimits? Images { get; init; }
 
+    /// <summary>
+    /// The role the system prompt is sent in, or null to leave it as it was.
+    /// </summary>
+    public SystemPromptRole? SystemPromptRole { get; init; }
+
     /// <summary>
     /// Applies this change to a profile.
     /// </summary>
@@ -80,5 +85,6 @@ public sealed record ModelProfileChange
         Context = this.Context ?? profile.Context,
         Tokenizer = this.Tokenizer ?? profile.Tokenizer,
         Images = this.Images ?? profile.Images,
+        SystemPromptRole = this.SystemPromptRole ?? profile.SystemPromptRole,
     };
 }

app/MindWork AI Studio/Models/ModelRuleBuilder.cs

@@ -33,6 +33,7 @@ public sealed class ModelRuleBuilder(string patternText, ModelRuleKind ruleKind,
     private ContextWindow? context;
     private TokenizerRef? tokenizer;
     private ImageLimits? images;
+    private SystemPromptRole? systemPromptRole;
 
     /// <summary>
     /// The text this rule answers for, before anything was stated about it.
@@ -226,6 +227,17 @@ public sealed class ModelRuleBuilder(string patternText, ModelRuleKind ruleKind,
         return this;
     }
 
+    /// <summary>
+    /// Which role the model takes its system prompt in.
+    /// </summary>
+    /// <param name="role">The role.</param>
+    /// <returns>The rule, to go on stating.</returns>
+    public ModelRuleBuilder SystemPromptRole(SystemPromptRole role)
+    {
+        this.systemPromptRole = role;
+        return this;
+    }
+
     /// <summary>
     /// Takes everything the rule stated before this one and goes on from there.
     /// </summary>
@@ -342,6 +354,7 @@ public sealed class ModelRuleBuilder(string patternText, ModelRuleKind ruleKind,
         Context = this.context ?? basis?.Context,
         Tokenizer = this.tokenizer ?? basis?.Tokenizer,
         Images = this.images ?? basis?.Images,
+        SystemPromptRole = this.systemPromptRole ?? basis?.SystemPromptRole,
     };
 
     private ModelRuleBuilder MatchingAs(MatchKind kind)

app/MindWork AI Studio/Models/OpenAI/Gpt35Family.cs

@@ -25,7 +25,8 @@ public sealed class Gpt35Family : ModelFamily
         builder.Rule("gpt-3.5").AsPrefix()
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT)
             .Apis(CHAT_COMPLETION_API)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base")
+            .SystemPromptRole(SystemPromptRole.SYSTEM);
 
         //
         // The odd one out, and kept odd on purpose: the previous rules put this one model on the
@@ -36,6 +37,7 @@ public sealed class Gpt35Family : ModelFamily
         builder.Rule("gpt-3.5-turbo").AsExact()
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT)
             .Apis(RESPONSES_API)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base")
+            .SystemPromptRole(SystemPromptRole.SYSTEM);
     }
 }

app/MindWork AI Studio/Models/OpenAI/Gpt4Family.cs

@@ -31,7 +31,8 @@ public sealed class Gpt4Family : ModelFamily
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT)
             .Apis(RESPONSES_API)
             .ContextWindow(8_192)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "cl100k_base")
+            .SystemPromptRole(SystemPromptRole.SYSTEM);
 
         builder.Rule("gpt-4-turbo").AsPrefix().Inherits()
             .Capabilities(MULTIPLE_IMAGE_INPUT | FUNCTION_CALLING)

app/MindWork AI Studio/Models/OpenAI/Gpt4oFamily.cs

@@ -33,7 +33,8 @@ public sealed class Gpt4oFamily : ModelFamily
             .Capabilities(TEXT_INPUT | MULTIPLE_IMAGE_INPUT | TEXT_OUTPUT | FUNCTION_CALLING | WEB_SEARCH)
             .Apis(RESPONSES_API)
             .ContextWindow(128_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
         //
         // The search previews are the same generation and almost nothing like it: they search the
@@ -45,12 +46,14 @@ public sealed class Gpt4oFamily : ModelFamily
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT | WEB_SEARCH)
             .Apis(CHAT_COMPLETION_API)
             .ContextWindow(128_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
         builder.Rule("gpt-4o-mini-search-preview").AsExact()
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT | WEB_SEARCH)
             .Apis(CHAT_COMPLETION_API)
             .ContextWindow(128_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
     }
 }

app/MindWork AI Studio/Models/OpenAI/Gpt5Family.cs

@@ -45,7 +45,8 @@ public sealed class Gpt5Family : ModelFamily
             .Apis(RESPONSES_API)
             .Reasoning(ReasoningSupport.ALWAYS)
             .ContextWindow(400_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
         //
         // The alias for the model of this generation which does not reason. The previous rules had

app/MindWork AI Studio/Models/OpenAI/Gpt6AstraFamily.cs

@@ -1,27 +0,0 @@
-using static AIStudio.Provider.Capability;
-
-namespace AIStudio.Models.OpenAI;
-
-/// <summary>
-/// GPT-6 Astra.
-/// </summary>
-/// <remarks>
-/// Unlike the 5.5 and 5.6 models it reasons on every request: the effort reaches from low to max,
-/// and there is no setting which switches thinking off.
-/// </remarks>
-public sealed class Gpt6AstraFamily : ModelFamily
-{
-    /// <inheritdoc />
-    public override ModelVendor Vendor => ModelVendor.OPEN_AI;
-
-    /// <inheritdoc />
-    public override ModelSource Source => new("https://developers.openai.com/api/docs/models", new DateOnly(2026, 9, 12), "Capabilities ported unchanged from ProviderExtensions.OpenAI.cs: reasons on every request, both APIs. The models page states the window as 1.05M tokens.");
-
-    /// <inheritdoc />
-    protected override void Declare(ModelFamilyBuilder builder) =>
-        builder.Rule("gpt-6-astra").AsPrefix()
-            .Capabilities(TEXT_INPUT | MULTIPLE_IMAGE_INPUT | TEXT_OUTPUT | FUNCTION_CALLING | WEB_SEARCH)
-            .Apis(RESPONSES_API | CHAT_COMPLETION_API)
-            .Reasoning(ReasoningSupport.ALWAYS)
-            .ContextWindow(1_050_000);
-}

app/MindWork AI Studio/Models/OpenAI/Gpt6Family.cs

@@ -0,0 +1,52 @@
+using static AIStudio.Provider.Capability;
+
+namespace AIStudio.Models.OpenAI;
+
+/// <summary>
+/// The whole GPT-6 line: Astra, Sol, and Luna.
+/// </summary>
+/// <remarks>
+/// One family for the generation rather than one per model, because the three differ in a single
+/// sentence: Astra reasons on every request -- its effort reaches from low to max and nothing
+/// switches thinking off -- while Sol and Luna reason unless the effort is set to none. Everything
+/// else, the window included, they share.
+///
+/// The rule for the generation is also what answers for a GPT-6 model OpenAI releases after this
+/// was written. That is on purpose: this family used to know Astra alone, and Sol and Luna went to
+/// the global assumption, which sent them to the chat completion API without images, web search,
+/// or reasoning.
+///
+/// Sol's page adds that calling functions through the chat completion API needs the effort set to
+/// none. That concerns only the gateways which reach these models through that API; OpenAI itself
+/// is asked through the Responses API.
+/// </remarks>
+public sealed class Gpt6Family : ModelFamily
+{
+    /// <inheritdoc />
+    public override ModelVendor Vendor => ModelVendor.OPEN_AI;
+
+    /// <inheritdoc />
+    public override ModelSource Source => new("https://developers.openai.com/api/docs/models", new DateOnly(2026, 9, 27), "The models page lists gpt-6-astra, gpt-6-sol, and gpt-6-luna as the GPT-6 line.");
+
+    /// <inheritdoc />
+    public override IReadOnlyList<ModelSource> FurtherSources =>
+    [
+        new("https://developers.openai.com/api/docs/models/gpt-6-astra", new DateOnly(2026, 9, 27), "Text and image input, text output, both APIs, function calling, and web search. A window of 1,050,000 tokens. The effort reaches from low to max, with no way to turn reasoning off."),
+        new("https://developers.openai.com/api/docs/models/gpt-6-sol", new DateOnly(2026, 9, 27), "The same abilities and window as Astra. The effort reaches from none to max and defaults to medium."),
+        new("https://developers.openai.com/api/docs/models/gpt-6-luna", new DateOnly(2026, 9, 27), "The same abilities and window as Astra. The effort reaches from none to max and defaults to medium.")
+    ];
+
+    /// <inheritdoc />
+    protected override void Declare(ModelFamilyBuilder builder)
+    {
+        builder.Rule("gpt-6").AsPrefix()
+            .Capabilities(TEXT_INPUT | MULTIPLE_IMAGE_INPUT | TEXT_OUTPUT | FUNCTION_CALLING | WEB_SEARCH)
+            .Apis(RESPONSES_API | CHAT_COMPLETION_API)
+            .Reasoning(ReasoningSupport.ON_BY_DEFAULT)
+            .ContextWindow(1_050_000)
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
+
+        builder.Rule("gpt-6-astra").AsPrefix().InheritsFrom("gpt-6")
+            .Reasoning(ReasoningSupport.ALWAYS);
+    }
+}

app/MindWork AI Studio/Models/OpenAI/GptBaselineFamily.cs

@@ -0,0 +1,45 @@
+using AIStudio.Provider;
+
+using static AIStudio.Provider.Capability;
+
+namespace AIStudio.Models.OpenAI;
+
+/// <summary>
+/// Every GPT model at OpenAI which no rule of its generation knows yet.
+/// </summary>
+/// <remarks>
+/// OpenAI releases faster than anybody writes rules, and somebody always reaches a new model on the
+/// day it appears. Unlike the global assumption, which has to stand for a hundred thousand models
+/// of every kind, this one stands for a single line in its vendor's own cloud, and that line is
+/// uniform: every GPT chat model OpenAI lists calls tools, looks at images, searches the web,
+/// reasons, and answers through the Responses API. Left to the global assumption instead, a new
+/// model went to the chat completion API without any of that, and a Responses API parameter the
+/// person had set was turned down -- which is what happened to GPT-6 Sol and Luna.
+///
+/// How it reasons is the one guess in here, and on by default is the guess which can be taken back:
+/// the person switches it off in the expert settings, whereas a model stated to always reason
+/// could not be told otherwise.
+///
+/// The rule is written as a name part on purpose. Specificity weighs how a pattern matches before
+/// it weighs how long it is, so a prefix rule for "gpt" would outrank the name part "gpt-oss" and
+/// answer for the open weights wherever they are served. As a name part, every longer rule beats
+/// it. Binding it to OpenAI keeps it away from the other models named that way, such as a
+/// self-hosted gpt-j. What the model is made for stays with the modifiers: an image, realtime, or
+/// transcription model is recognized by its name as before.
+/// </remarks>
+public sealed class GptBaselineFamily : ModelFamily
+{
+    /// <inheritdoc />
+    public override ModelVendor Vendor => ModelVendor.OPEN_AI;
+
+    /// <inheritdoc />
+    public override ModelSource Source => new("https://developers.openai.com/api/docs/models", new DateOnly(2026, 9, 27), "Every GPT chat model the page lists supports function calling, image input, web search, reasoning, and the Responses API. The models which are no chat models carry image, realtime, live, transcribe, or tts in their names.");
+
+    /// <inheritdoc />
+    protected override void Declare(ModelFamilyBuilder builder) =>
+        builder.Rule("gpt").AsSegment().OnlyOn(LLMProviders.OPEN_AI)
+            .Capabilities(TEXT_INPUT | MULTIPLE_IMAGE_INPUT | TEXT_OUTPUT | FUNCTION_CALLING | WEB_SEARCH)
+            .Apis(RESPONSES_API)
+            .Reasoning(ReasoningSupport.ON_BY_DEFAULT)
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
+}

app/MindWork AI Studio/Models/OpenAI/OSeriesFamily.cs

@@ -36,26 +36,38 @@ public sealed class OSeriesFamily : ModelFamily
             .Apis(RESPONSES_API)
             .Reasoning(ReasoningSupport.ALWAYS)
             .ContextWindow(200_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
+        //
+        // The first mini and the preview came out before OpenAI let a reasoning model take
+        // instructions of its own, so whatever the system prompt says has to reach them as a
+        // message of the person.
+        //
         builder.Rule("o1-mini").AsPrefix()
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT)
             .Apis(CHAT_COMPLETION_API)
             .Reasoning(ReasoningSupport.ALWAYS)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.USER);
+
+        builder.Rule("o1-preview").AsPrefix().InheritsFrom("o1")
+            .SystemPromptRole(SystemPromptRole.USER);
 
         builder.Rule("o3").AsPrefix()
             .Capabilities(TEXT_INPUT | MULTIPLE_IMAGE_INPUT | TEXT_OUTPUT | FUNCTION_CALLING | WEB_SEARCH)
             .Apis(RESPONSES_API)
             .Reasoning(ReasoningSupport.ALWAYS)
             .ContextWindow(200_000)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
         builder.Rule("o3-mini").AsPrefix()
             .Capabilities(TEXT_INPUT | TEXT_OUTPUT | FUNCTION_CALLING)
             .Apis(RESPONSES_API)
             .Reasoning(ReasoningSupport.ALWAYS)
-            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+            .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+            .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
         // The one mini which is not cut down: it is the o3 generation under another number.
         builder.Rule("o4-mini").AsPrefix().InheritsFrom("o3");

app/MindWork AI Studio/Models/SystemPromptRole.cs

@@ -0,0 +1,36 @@
+namespace AIStudio.Models;
+
+/// <summary>
+/// Which role the system prompt is sent in.
+/// </summary>
+/// <remarks>
+/// A question only OpenAI's own cloud asks. Its models took their instructions as "system" first,
+/// then as "developer", and the early reasoning models took none at all, so that the instructions
+/// had to travel as a message of the person instead. Every other provider has one way to send a
+/// system prompt, and for their models the rules say nothing about this.
+///
+/// That the vendor's name is not in the members is on purpose: they say what the model accepts,
+/// and the provider decides how to spell that in its request.
+/// </remarks>
+public enum SystemPromptRole
+{
+    /// <summary>
+    /// No rule says anything about it, so the provider decides.
+    /// </summary>
+    UNKNOWN,
+
+    /// <summary>
+    /// The model takes its instructions in the system role.
+    /// </summary>
+    SYSTEM,
+
+    /// <summary>
+    /// The model takes its instructions in the developer role.
+    /// </summary>
+    DEVELOPER,
+
+    /// <summary>
+    /// The model takes no instructions of its own, so they go as a message of the person.
+    /// </summary>
+    USER,
+}

app/MindWork AI Studio/Provider/OpenAI/ProviderOpenAI.cs

@@ -63,38 +63,16 @@ public sealed class ProviderOpenAI() : BaseProvider(LLMProviders.OPEN_AI, new Ur
         if(!requestedSecret.Success)
             yield break;
         
-        // Unfortunately, OpenAI changed the name of the system prompt based on the model.
-        // All models that start with "o" (the omni aka reasoning models), all GPT4o models,
-        // and all newer models have the system prompt named "developer". All other models
-        // have the system prompt named "system". We need to check this to get the correct
-        // system prompt.
-        //
-        // To complicate it even more: The early versions of reasoning models, which are released
-        // before the 17th of December 2024, have no system prompt at all. We need to check this
-        // as well.
-        
-        // Apply the basic rule first:
-        var systemPromptRole =
-            chatModel.Id.StartsWith('o') ||
-            chatModel.Id.StartsWith("gpt-5", StringComparison.Ordinal) ||
-            chatModel.Id.Contains("4o") ? "developer" : "system";
-        
-        // Check if the model is an early version of the reasoning models:
-        systemPromptRole = chatModel.Id switch
-        {
-            "o1-mini" => "user",
-            "o1-mini-2024-09-12" => "user",
-            "o1-preview" => "user",
-            "o1-preview-2024-09-12" => "user",
-            
-            _ => systemPromptRole,
-        };
-
         // Read the model capabilities. Through the settings provider, so that the user's expert
         // capability overrides apply:
         var providerSettings = this.CreateSettingsProvider(chatModel);
         var modelProfile = providerSettings.GetModelProfile();
-        
+
+        // OpenAI changed the role of the system prompt from one generation to the next, and the
+        // early reasoning models take no system prompt at all. Which role a model takes is stated
+        // by the OpenAI families in Models/OpenAI:
+        var systemPromptRole = modelProfile.SystemPromptRole.ToOpenAIRole();
+
         // Check if we are using the Responses API or the Chat Completion API:
         var usingResponsesAPI = modelProfile.Has(Capability.RESPONSES_API);
         

app/MindWork AI Studio/Provider/OpenAI/SystemPromptRoleExtensions.cs

@@ -0,0 +1,25 @@
+using AIStudio.Models;
+
+namespace AIStudio.Provider.OpenAI;
+
+public static class SystemPromptRoleExtensions
+{
+    /// <summary>
+    /// Spells a system prompt role the way OpenAI's APIs expect it.
+    /// </summary>
+    /// <remarks>
+    /// A role no rule states becomes "developer". OpenAI's current documentation names no other, a
+    /// model the rules do not know is almost always one released after they were written, and the
+    /// older generations which still want "system" say so in their families.
+    /// </remarks>
+    /// <param name="role">The role the model profile states.</param>
+    /// <returns>The role, as OpenAI names it.</returns>
+    public static string ToOpenAIRole(this SystemPromptRole role) => role switch
+    {
+        SystemPromptRole.SYSTEM => "system",
+        SystemPromptRole.DEVELOPER => "developer",
+        SystemPromptRole.USER => "user",
+
+        _ => "developer",
+    };
+}

app/MindWork AI Studio/wwwroot/changelog/v26.9.1.md

@@ -17,7 +17,7 @@
 - Added a Search Confluence tool, so the AI can look things up in your organization's Confluence Data Center wiki. Set your wiki's address in the tool settings. Selecting the tool also selects Read Web Page, which the AI uses to open the pages it found. Confluence Cloud is not supported yet. Many thanks to Peer Hogeterp (`peerschuett`) for this tool.
 - Added sign-in with your operating system account to the Search Confluence tool for a wiki on your organization's network, so you do not have to enter a password. When your wiki does not accept that sign-in, AI Studio tells you so instead of reporting an empty search.
 - Added safeguards to the Search Confluence tool: it works only with a High-confidence provider, and it never follows a redirect that leads away from your wiki. Each search appears in the sources of the answer.
-- Added support for OpenAI's GPT-6 Astra.
+- Added support for OpenAI's GPT-6 Astra, Sol, and Luna.
 - Added a way to keep improving a prompt in the Prompt Optimizer. Select "Improve further" to move the latest proposal back into the prompt field, edit it the way you want, and optimize it again. Your recommendations and everything you selected stay as they are.
 - Added the context window to what AI Studio knows about a model, wherever its metadata states one.
 - Added a live read of that context window at the providers which report it, among them Mistral, Groq, OpenRouter, and self-hosted vLLM servers. You then get the window your own server was started with, not the one the model card advertises.

app/Tests/Models/CapabilityCharacterizationTests.cs

@@ -27,11 +27,12 @@ public sealed class CapabilityCharacterizationTests
     /// How many columns a snapshot line carries after the model ID.
     /// </summary>
     /// <remarks>
-    /// The capabilities, the kind, the context window, the image limit, and the tokenizer. Adding a
-    /// column to the snapshot means raising this, and forgetting to would split a line inside its
-    /// last column instead of in front of it -- which makes every model look changed at once.
+    /// The capabilities, the kind, the context window, the image limit, the tokenizer, and the system
+    /// prompt role. Adding a column to the snapshot means raising this, and forgetting to would split
+    /// a line inside its last column instead of in front of it -- which makes every model look
+    /// changed at once.
     /// </remarks>
-    private const int TRAILING_COLUMNS = 5;
+    private const int TRAILING_COLUMNS = 6;
 
     [Test]
     public void TheCorpusStillGetsTheAnswersTheSnapshotRecorded()

app/Tests/Models/ContextWindowRuleTests.cs

@@ -27,6 +27,8 @@ public sealed class ContextWindowRuleTests
     [TestCase(LLMProviders.OPEN_AI, "gpt-5.5", 1_050_000)]
     [TestCase(LLMProviders.OPEN_AI, "gpt-5.6", 1_050_000)]
     [TestCase(LLMProviders.OPEN_AI, "gpt-6-astra", 1_050_000)]
+    [TestCase(LLMProviders.OPEN_AI, "gpt-6-sol", 1_050_000)]
+    [TestCase(LLMProviders.OPEN_AI, "gpt-6-luna", 1_050_000)]
     [TestCase(LLMProviders.OPEN_AI, "o1", 200_000)]
     [TestCase(LLMProviders.OPEN_AI, "o3", 200_000)]
     [TestCase(LLMProviders.OPEN_AI, "o4-mini", 200_000, Description = "The o3 generation under another number, window included.")]

app/Tests/Models/Corpus/CapabilitySnapshot.cs

@@ -57,6 +57,15 @@ public static class CapabilitySnapshot
     /// </remarks>
     private const string NO_TOKENIZER = "(unknown)";
 
+    /// <summary>
+    /// What a model no rule states a system prompt role for is written as.
+    /// </summary>
+    /// <remarks>
+    /// Which is nearly every model outside OpenAI's own line, and rightly so: the provider then
+    /// sends the system prompt the one way it knows.
+    /// </remarks>
+    private const string NO_SYSTEM_PROMPT_ROLE = "(unknown)";
+
     private const string HEADER =
         """
         # What the rules answer, for every model of the corpus.
@@ -66,8 +75,11 @@ public static class CapabilitySnapshot
         #
         # Columns are provider, model ID as the provider reports it, the capabilities sorted by
         # name, what the model is made for, its context window in tokens, how many images it
-        # accepts, and which tokenizer it uses. The ID stands here unchanged, so a line may well
-        # carry leading or trailing spaces.
+        # accepts, which tokenizer it uses, and which role its system prompt is sent in. The ID
+        # stands here unchanged, so a line may well carry leading or trailing spaces.
+        #
+        # A system prompt role written as "(unknown)" leaves the choice to the provider. Only the
+        # rules of OpenAI's own line state one, because only OpenAI's cloud asks the question.
         #
         # A window or an image limit written as "(unknown)" is one nobody has stated a source for.
         # That is a gap, not a claim: the app then shows a person how many tokens their conversation
@@ -137,9 +149,16 @@ public static class CapabilitySnapshot
     private static string Line(CorpusEntry entry)
     {
         var profile = entry.Provider.GetModelProfile(new Model(entry.ModelId, null));
-        return $"{entry.Provider} | {entry.ModelId} | {Describe(RebuiltRules.AsCapabilities(profile))} | {profile.Kind} | {Describe(profile.Context)} | {Describe(profile.Images)} | {Describe(profile.Tokenizer)}";
+        return $"{entry.Provider} | {entry.ModelId} | {Describe(RebuiltRules.AsCapabilities(profile))} | {profile.Kind} | {Describe(profile.Context)} | {Describe(profile.Images)} | {Describe(profile.Tokenizer)} | {Describe(profile.SystemPromptRole)}";
     }
 
+    /// <summary>
+    /// Writes a system prompt role the way a snapshot line does.
+    /// </summary>
+    /// <param name="role">The role to write.</param>
+    /// <returns>The role, or a marker when no rule stated one.</returns>
+    public static string Describe(SystemPromptRole role) => role is SystemPromptRole.UNKNOWN ? NO_SYSTEM_PROMPT_ROLE : role.ToString();
+
     /// <summary>
     /// Writes a tokenizer reference the way a snapshot line does.
     /// </summary>

app/Tests/Models/Corpus/CapabilitySnapshot.txt

@@ -5,8 +5,11 @@
 #
 # Columns are provider, model ID as the provider reports it, the capabilities sorted by
 # name, what the model is made for, its context window in tokens, how many images it
-# accepts, and which tokenizer it uses. The ID stands here unchanged, so a line may well
-# carry leading or trailing spaces.
+# accepts, which tokenizer it uses, and which role its system prompt is sent in. The ID
+# stands here unchanged, so a line may well carry leading or trailing spaces.
+#
+# A system prompt role written as "(unknown)" leaves the choice to the provider. Only the
+# rules of OpenAI's own line state one, because only OpenAI's cloud asks the question.
 #
 # A window or an image limit written as "(unknown)" is one nobody has stated a source for.
 # That is a gap, not a claim: the app then shows a person how many tokens their conversation
@@ -19,278 +22,281 @@
 # corrected, so keeping them out only hid four of their columns: ExpectedChanges.cs states
 # what each of them must answer, but it states capabilities alone.
 #
-ALIBABA_CLOUD | qvq-max | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen-mt-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen-vl-max | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen2.5-14b-instruct-1m | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen2.5-72b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen2.5-omni-7b | AUDIO_INPUT, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen2.5-vl-72b-instruct | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3-omni-flash | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3-vl-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.5-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.6-max | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.7-max | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.7-max-2026-05-17 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.7-max-2026-06-08 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.7-max-preview | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.8-27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.8-flash | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwen3.8-max | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | (unknown) | (unknown)
-ALIBABA_CLOUD | qwq-32b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | qwq-plus | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-ALIBABA_CLOUD | text-embedding-v3 | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown)
-ANTHROPIC | claude-3-5-haiku-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-3-5-sonnet-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-3-7-sonnet-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-3-opus-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-7-sonnet | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-fable-5-1 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-haiku-4-5-20251001 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-mythos-5 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-opus-4-0 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-opus-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-sonnet-4-0 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens
-ANTHROPIC | claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-DEEP_SEEK | deepseek-chat | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-DEEP_SEEK | deepseek-reasoner | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-DEEP_SEEK | deepseek-v3.2-exp | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-DEEP_SEEK | deepseek-v4 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown)
-DEEP_SEEK | deepseek-v4-vision | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown)
-FIREWORKS | accounts/fireworks/models/deepseek-v3 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-FIREWORKS | accounts/fireworks/models/llama-v3p1-405b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-FIREWORKS | accounts/fireworks/models/qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-FIREWORKS | whisper-v3 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-1.0-pro-vision | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-2.0-flash | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-2.0-flash-live-001 | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | REALTIME | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-2.5-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-2.5-flash-image | CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-2.5-flash-lite | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-2.5-pro | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-3-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-3-pro | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-3-pro-image | ALWAYS_REASONING, CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-3.1-flash-image | ALWAYS_REASONING, CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown)
-GOOGLE | gemini-flash-latest | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | gemini-pro-latest | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens
-GOOGLE | imagen-4.0-generate-001 | IMAGE_OUTPUT, TEXT_INPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown)
-GOOGLE | text-embedding-004 | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown)
-GROQ | llama-3.3-70b-versatile | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-GROQ | moonshotai/kimi-k2-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GROQ | openai/gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown)
-GROQ | qwen/qwen3-32b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GROQ | whisper-large-v3-turbo | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-GWDG | claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-GWDG | deepseek-r1 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GWDG | e5-mistral-7b-instruct | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown)
-GWDG | gemma-3-27b-it | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GWDG | gpt-5.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-GWDG | internvl2.5-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GWDG | meta-llama-3.1-8b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-GWDG | qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-GWDG | whisper-large-v2 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-HELMHOLTZ | 01 - GPT-5.5 - great overall performance | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-HELMHOLTZ | 1 - Llama3 405 the best general model | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HELMHOLTZ | 10 - Muse Glimmer 30b - the newest META model | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HELMHOLTZ | Qwen 3.8-27B with DFlash on haicluster | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HELMHOLTZ | alias-qwen38-27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HETZNER | gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown)
-HETZNER | qwen3-coder-30b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | HuggingFaceTB/SmolLM3-3B | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | Qwen/Qwen3.8-27B | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | google/gemma-4-31B-it:novita | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | meta-llama/Llama-4-Scout-17B-16E-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | meta-llama/Meta-Llama-3.1-405B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-HUGGINGFACE | mistralai/Magistral-Small-2509 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-HUGGINGFACE | openai/gpt-oss-120b:fireworks-ai | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown)
-IONOS | meta-llama/Llama-3.3-70B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-IONOS | mistralai/Mistral-Small-24B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-LITE_LLM | anthropic/claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-LITE_LLM | azure/gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-LITE_LLM | bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-LITE_LLM | the-fast-one | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | codestral-2508 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | TEXT_COMPLETION | (unknown) | (unknown) | (unknown)
-MISTRAL | magistral-medium-2506 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | ministral-14b-2512 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | ministral-3b-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | ministral-8b-2410 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | mistral-large-2411 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-large-2512 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-large-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-2505 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-2508 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-2604 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-3-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-3.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-medium-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-MISTRAL | mistral-saba-2502 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | mistral-small-2501 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | mistral-small-2503 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | mistral-small-2603 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | mistral-small-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | open-mistral-nemo | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-MISTRAL | pixtral-12b-2409 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | (unknown)
-MISTRAL | pixtral-large-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | (unknown)
-MISTRAL | voxtral-small-2507 | CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-NONE | gpt-5.6 | (nothing) | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_AI |  | (nothing) | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_AI | gpt-3.5-turbo | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | gpt-3.5-turbo-16k | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | gpt-4 | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 8192 | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | gpt-4-0613 | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 8192 | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | gpt-4-turbo | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | gpt-4o | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-4o-audio-preview | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | SPEECH_SYNTHESIS | 128000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-4o-mini | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-4o-mini-search-preview | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-4o-search-preview | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5-chat-latest | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5-mini | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5-nano | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.1-codex | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.2 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.3 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.4 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | gpt-6-astra | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown)
-OPEN_AI | gpt-6-astra-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown)
-OPEN_AI | o1 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o1-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o1-pro | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o3 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o3-mini | ALWAYS_REASONING, FUNCTION_CALLING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o3-pro | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | o4-mini | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base
-OPEN_AI | text-embedding-3-large | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | TIKTOKEN cl100k_base
-OPEN_AI | whisper-1 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | anthropic/claude-opus-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens
-OPEN_ROUTER | deepseek/deepseek-chat-v3.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | deepseek/deepseek-r1-distill-llama-70b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | google/gemini-3.7-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens
-OPEN_ROUTER | google/gemma-4-31b-it | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | meta-llama/llama-4-maverick | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | minimax/minimax-m2 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | mistralai/mistral-large-3 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
-OPEN_ROUTER | moonshotai/kimi-k2-thinking | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | nvidia/nemotron-3-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | openai/gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base
-OPEN_ROUTER | openai/gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown)
-OPEN_ROUTER | perplexity/sonar-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | qwen/qwen3.8-flash-next | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-OPEN_ROUTER | z-ai/glm-5.3 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-PERPLEXITY | sonar | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-PERPLEXITY | sonar-deep-research | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-PERPLEXITY | sonar-pro | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-PERPLEXITY | sonar-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-PERPLEXITY | sonar-reasoning-pro | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED |     | (nothing) | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | --- | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | 01-ai/yi-large | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | a-model-nobody-has-heard-of | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | apertus-1.5-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | apriel-1.5-15b-thinker | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | apriel-1.6-15b-thinker | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | aya-expanse:8b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | aya-vision:8b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | command-a-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | command-a-reasoning | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | command-a-vision | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | command-a:111b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | command-r7b:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | deepseek-r1-distill-llama-70b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | deepseek-r1:32b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | deepseek-v2.5 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | deepseek-v3.1:671b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ernie-4.5-21b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ernie-4.5-vl-28b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ernie-x1.1-thinking | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | eurollm-9b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | falcon-h1-1.5b-tool-calling | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | falcon-h1:7b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | falcon3:10b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma2:9b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma3:1b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma3:27b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma3n:e4b | AUDIO_INPUT, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma4:31b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gemma4:e2b | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | glm-4-9b-chat | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | glm-4.5v | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | glm-4.6:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | glm-5-2 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | glm-5.3-flash-nvfp4 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | gpt-oss:20b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown)
-SELF_HOSTED | granite-embedding:278m | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | granite3.2-vision:2b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | granite3.3:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | granite4.2:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | hunyuan:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | inclusionai/ling-mini-2.0 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | internlm3:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | internvl3-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | kimi-k2.7-code | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | kimi-k2:1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | kimi-k3:latest | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | kimi-vl:16b | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ling-1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | llama-3.1-405b-base | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-SELF_HOSTED | llama-3.3-nemotron-super-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | llama2:13b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | llama3.2-vision:11b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | llama3.2:3b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown)
-SELF_HOSTED | magistral:24b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | minimax-m2:latest | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | minimax-text-01 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ministral-8b-instruct-2410 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | mistral-nemo:12b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | mistral-small-3.1-24b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | mistral-small3.2:24b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | muse-glimmer-30b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | nemotron-3-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | nomic-embed-text:latest | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | nvidia-nemotron-3.5-lightning-30b-a3b-nvfp4 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | occiglot-7b-eu5 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | olmo-3-32b-think | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | olmo2:13b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | olmo3:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi-4-mini-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi-4-multimodal-instruct | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi-4-reasoning-vision | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi3:14b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi4-mini:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | phi4:14b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen2.5-vl-7b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3-coder:30b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.5:32b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.6:32b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.8-2.4t-a95b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.8:27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.8:27b-mlx | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwen3.8:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | qwq:32b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | ring-1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | salamandra-7b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | salamandra-7b-instruct-tools | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | seed-oss:36b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | smollm2:1.7b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | smollm3:3b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | starling-lm:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | tencent/hy3 | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | teuken-7b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | voxtral-mini-3b | CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown)
-SELF_HOSTED | yi-1.5:9b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-2-vision-1212 | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-3 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-3-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-4 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-4-fast-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-4.20 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown)
-X | grok-4.20-non-reasoning | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown)
-X | grok-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown)
-X | grok-build-0.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown)
+ALIBABA_CLOUD | qvq-max | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen-mt-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen-vl-max | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen2.5-14b-instruct-1m | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen2.5-72b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen2.5-omni-7b | AUDIO_INPUT, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen2.5-vl-72b-instruct | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3-omni-flash | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3-vl-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.5-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.6-max | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.7-max | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.7-max-2026-05-17 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.7-max-2026-06-08 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.7-max-preview | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.8-27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.8-flash | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwen3.8-max | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwq-32b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | qwq-plus | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+ALIBABA_CLOUD | text-embedding-v3 | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown) | (unknown)
+ANTHROPIC | claude-3-5-haiku-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-3-5-sonnet-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-3-7-sonnet-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-3-opus-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-7-sonnet | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-fable-5-1 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-haiku-4-5-20251001 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-mythos-5 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-opus-4-0 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-opus-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-sonnet-4-0 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | 100 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+ANTHROPIC | claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+DEEP_SEEK | deepseek-chat | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+DEEP_SEEK | deepseek-reasoner | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+DEEP_SEEK | deepseek-v3.2-exp | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+DEEP_SEEK | deepseek-v4 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown) | (unknown)
+DEEP_SEEK | deepseek-v4-vision | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown) | (unknown)
+FIREWORKS | accounts/fireworks/models/deepseek-v3 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+FIREWORKS | accounts/fireworks/models/llama-v3p1-405b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+FIREWORKS | accounts/fireworks/models/qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+FIREWORKS | whisper-v3 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-1.0-pro-vision | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-2.0-flash | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-2.0-flash-live-001 | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, SPEECH_OUTPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | REALTIME | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-2.5-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-2.5-flash-image | CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-2.5-flash-lite | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-2.5-pro | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1048576 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-3-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-3-pro | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-3-pro-image | ALWAYS_REASONING, CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-3.1-flash-image | ALWAYS_REASONING, CHAT_COMPLETION_API, IMAGE_OUTPUT, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | gemini-flash-latest | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | gemini-pro-latest | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens | (unknown)
+GOOGLE | imagen-4.0-generate-001 | IMAGE_OUTPUT, TEXT_INPUT | IMAGE_GENERATION | (unknown) | (unknown) | (unknown) | (unknown)
+GOOGLE | text-embedding-004 | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown) | (unknown)
+GROQ | llama-3.3-70b-versatile | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+GROQ | moonshotai/kimi-k2-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GROQ | openai/gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+GROQ | qwen/qwen3-32b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GROQ | whisper-large-v3-turbo | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+GWDG | deepseek-r1 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | e5-mistral-7b-instruct | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | gemma-3-27b-it | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | gpt-5.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+GWDG | internvl2.5-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | meta-llama-3.1-8b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+GWDG | qwen3-235b-a22b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+GWDG | whisper-large-v2 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+HELMHOLTZ | 01 - GPT-5.5 - great overall performance | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+HELMHOLTZ | 1 - Llama3 405 the best general model | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HELMHOLTZ | 10 - Muse Glimmer 30b - the newest META model | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HELMHOLTZ | Qwen 3.8-27B with DFlash on haicluster | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HELMHOLTZ | alias-qwen38-27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HETZNER | gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+HETZNER | qwen3-coder-30b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | HuggingFaceTB/SmolLM3-3B | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | Qwen/Qwen3.8-27B | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | deepseek-ai/DeepSeek-R1-Distill-Qwen-32B | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | google/gemma-4-31B-it:novita | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | meta-llama/Llama-4-Scout-17B-16E-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | meta-llama/Meta-Llama-3.1-405B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | mistralai/Magistral-Small-2509 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+HUGGINGFACE | openai/gpt-oss-120b:fireworks-ai | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+IONOS | meta-llama/Llama-3.3-70B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+IONOS | mistralai/Mistral-Small-24B-Instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+LITE_LLM | anthropic/claude-sonnet-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+LITE_LLM | azure/gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+LITE_LLM | bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+LITE_LLM | the-fast-one | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | codestral-2508 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | TEXT_COMPLETION | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | magistral-medium-2506 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | ministral-14b-2512 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | ministral-3b-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | ministral-8b-2410 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-large-2411 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-large-2512 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-large-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-2505 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-2508 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-2604 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-3-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-3.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-medium-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-saba-2502 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-small-2501 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-small-2503 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-small-2603 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | mistral-small-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | open-mistral-nemo | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+MISTRAL | pixtral-12b-2409 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | (unknown) | (unknown)
+MISTRAL | pixtral-large-latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | (unknown) | (unknown)
+MISTRAL | voxtral-small-2507 | CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+NONE | gpt-5.6 | (nothing) | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_AI |  | (nothing) | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_AI | gpt-3.5-turbo | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN cl100k_base | SYSTEM
+OPEN_AI | gpt-3.5-turbo-16k | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN cl100k_base | SYSTEM
+OPEN_AI | gpt-4 | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 8192 | (unknown) | TIKTOKEN cl100k_base | SYSTEM
+OPEN_AI | gpt-4-0613 | RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 8192 | (unknown) | TIKTOKEN cl100k_base | SYSTEM
+OPEN_AI | gpt-4-turbo | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 128000 | (unknown) | TIKTOKEN cl100k_base | SYSTEM
+OPEN_AI | gpt-4o | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-4o-audio-preview | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | SPEECH_SYNTHESIS | 128000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-4o-mini | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-4o-mini-search-preview | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-4o-search-preview | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 128000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5-chat-latest | FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5-mini | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5-nano | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.1-codex | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.2 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 400000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.3 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.4 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | gpt-6-astra | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown) | DEVELOPER
+OPEN_AI | gpt-6-astra-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown) | DEVELOPER
+OPEN_AI | gpt-6-luna | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown) | DEVELOPER
+OPEN_AI | gpt-6-sol | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | (unknown) | DEVELOPER
+OPEN_AI | o1 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | o1-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base | USER
+OPEN_AI | o1-preview | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | USER
+OPEN_AI | o1-pro | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | o3 | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | o3-mini | ALWAYS_REASONING, FUNCTION_CALLING, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | o3-pro | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | o4-mini | ALWAYS_REASONING, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, RESPONSES_API, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 200000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_AI | text-embedding-3-large | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | TIKTOKEN cl100k_base | (unknown)
+OPEN_AI | whisper-1 | SPEECH_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | anthropic/claude-opus-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | 600 per request | PROVIDER_API /v1/messages/count_tokens | (unknown)
+OPEN_ROUTER | deepseek/deepseek-chat-v3.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | deepseek/deepseek-r1-distill-llama-70b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | google/gemini-3.7-flash | ALWAYS_REASONING, AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | 1000000 | 3600 per request | PROVIDER_API countTokens | (unknown)
+OPEN_ROUTER | google/gemma-4-31b-it | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | meta-llama/llama-4-maverick | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | minimax/minimax-m2 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | mistralai/mistral-large-3 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | moonshotai/kimi-k2-thinking | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | nvidia/nemotron-3-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | openai/gpt-5.6 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 1050000 | (unknown) | TIKTOKEN o200k_base | DEVELOPER
+OPEN_ROUTER | openai/gpt-oss-120b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | perplexity/sonar-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | qwen/qwen3.8-flash-next | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+OPEN_ROUTER | z-ai/glm-5.3 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+PERPLEXITY | sonar | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+PERPLEXITY | sonar-deep-research | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+PERPLEXITY | sonar-pro | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+PERPLEXITY | sonar-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+PERPLEXITY | sonar-reasoning-pro | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED |     | (nothing) | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | --- | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | 01-ai/yi-large | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | a-model-nobody-has-heard-of | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | apertus-1.5-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | apriel-1.5-15b-thinker | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | apriel-1.6-15b-thinker | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | aya-expanse:8b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | aya-vision:8b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | command-a-plus | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | command-a-reasoning | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | command-a-vision | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | command-a:111b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | command-r7b:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | deepseek-r1-distill-llama-70b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | deepseek-r1:32b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | deepseek-v2.5 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | deepseek-v3.1:671b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ernie-4.5-21b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ernie-4.5-vl-28b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ernie-x1.1-thinking | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | eurollm-9b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | falcon-h1-1.5b-tool-calling | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | falcon-h1:7b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | falcon3:10b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma2:9b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma3:1b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma3:27b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma3n:e4b | AUDIO_INPUT, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma4:31b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gemma4:e2b | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | glm-4-9b-chat | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | glm-4.5v | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | glm-4.6:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | glm-5-2 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | glm-5.3-flash-nvfp4 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | gpt-oss:20b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT, WEB_SEARCH | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | granite-embedding:278m | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | granite3.2-vision:2b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | granite3.3:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | granite4.2:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | hunyuan:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | inclusionai/ling-mini-2.0 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | internlm3:8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | internvl3-8b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | kimi-k2.7-code | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | kimi-k2:1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | kimi-k3:latest | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT, VIDEO_INPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | kimi-vl:16b | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ling-1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | llama-3.1-405b-base | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | llama-3.3-nemotron-super-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | llama2:13b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | llama3.2-vision:11b | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | llama3.2:3b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | 131072 | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | magistral:24b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | minimax-m2:latest | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | minimax-text-01 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ministral-8b-instruct-2410 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | mistral-nemo:12b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | mistral-small-3.1-24b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | mistral-small3.2:24b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | muse-glimmer-30b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | nemotron-3-49b | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | nomic-embed-text:latest | EMBEDDING, TEXT_INPUT | EMBEDDING | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | nvidia-nemotron-3.5-lightning-30b-a3b-nvfp4 | CHAT_COMPLETION_API, FUNCTION_CALLING, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | occiglot-7b-eu5 | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | olmo-3-32b-think | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | olmo2:13b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | olmo3:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi-4-mini-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi-4-multimodal-instruct | AUDIO_INPUT, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi-4-reasoning-vision | ALWAYS_REASONING, CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi3:14b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi4-mini:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | phi4:14b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen2.5-vl-7b-instruct | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3-coder:30b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.5:32b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.6:32b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.8-2.4t-a95b | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.8:27b | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.8:27b-mlx | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwen3.8:latest | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, REASONING_BY_DEFAULT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | qwq:32b | ALWAYS_REASONING, CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | ring-1t | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | salamandra-7b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | salamandra-7b-instruct-tools | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | seed-oss:36b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | smollm2:1.7b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | smollm3:3b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | starling-lm:7b | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | tencent/hy3 | CHAT_COMPLETION_API, FUNCTION_CALLING, OPTIONAL_REASONING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | teuken-7b-instruct | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | voxtral-mini-3b | CHAT_COMPLETION_API, FUNCTION_CALLING, SPEECH_INPUT, TEXT_INPUT, TEXT_OUTPUT | TRANSCRIPTION | (unknown) | (unknown) | (unknown) | (unknown)
+SELF_HOSTED | yi-1.5:9b | CHAT_COMPLETION_API, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-2-vision-1212 | CHAT_COMPLETION_API, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-3 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-3-mini | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-4 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-4-fast-reasoning | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-4.20 | ALWAYS_REASONING, CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown) | (unknown)
+X | grok-4.20-non-reasoning | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 1000000 | (unknown) | (unknown) | (unknown)
+X | grok-5 | CHAT_COMPLETION_API, FUNCTION_CALLING, TEXT_INPUT, TEXT_OUTPUT | CHAT | (unknown) | (unknown) | (unknown) | (unknown)
+X | grok-build-0.1 | CHAT_COMPLETION_API, FUNCTION_CALLING, MULTIPLE_IMAGE_INPUT, TEXT_INPUT, TEXT_OUTPUT | CHAT | 256000 | (unknown) | (unknown) | (unknown)

app/Tests/Models/Corpus/ModelCorpus.cs

@@ -29,6 +29,8 @@ public static class ModelCorpus
     [
         new(OPEN_AI, "gpt-6-astra", NAMED_BY_A_RULE),
         new(OPEN_AI, "gpt-6-astra-mini", NAMED_BY_A_RULE),
+        new(OPEN_AI, "gpt-6-sol", NAMED_BY_NO_RULE),
+        new(OPEN_AI, "gpt-6-luna", NAMED_BY_NO_RULE),
         new(OPEN_AI, "gpt-5.6", NAMED_BY_A_RULE),
         new(OPEN_AI, "gpt-5.5", ON_THE_MANUAL_TEST_LIST),
         new(OPEN_AI, "gpt-5.4", NAMED_BY_A_RULE),
@@ -53,6 +55,7 @@ public static class ModelCorpus
         new(OPEN_AI, "o1", NAMED_BY_A_RULE),
         new(OPEN_AI, "o1-pro", NAMED_BY_A_RULE),
         new(OPEN_AI, "o1-mini", NAMED_BY_A_RULE),
+        new(OPEN_AI, "o1-preview", NAMED_BY_A_RULE),
         new(OPEN_AI, "o3", NAMED_BY_A_RULE),
         new(OPEN_AI, "o3-pro", NAMED_BY_A_RULE),
         new(OPEN_AI, "o3-mini", NAMED_BY_A_RULE),

app/Tests/Models/ModelFactsTests.cs

@@ -3,7 +3,7 @@ using AIStudio.Models;
 namespace AIStudio.Tests.Models;
 
 /// <summary>
-/// Checks the three types which have to be able to say "nobody knows".
+/// Checks the four types which have to be able to say "nobody knows".
 /// </summary>
 /// <remarks>
 /// They are tested together because they are tested for the same thing. Each of them is a value
@@ -98,4 +98,16 @@ public sealed class ModelFactsTests
             Assert.That(noImages.MaxPerMessage, Is.EqualTo(0));
         });
     }
+
+    [Test]
+    public void ASystemPromptRoleNobodyWroteDownIsUnknown()
+    {
+        //
+        // Not the system role, although that is the oldest one: an unknown role is what lets the
+        // provider decide, and a default reading as a statement would take that away.
+        //
+        SystemPromptRole untouched = default;
+
+        Assert.That(untouched, Is.EqualTo(SystemPromptRole.UNKNOWN));
+    }
 }

app/Tests/Models/ModelFamilyTests.cs

@@ -59,6 +59,20 @@ public sealed class ModelFamilyTests
             Assert.That(codex.Reasoning, Is.EqualTo(ReasoningSupport.OPTIONAL));
             Assert.That(codex.Context.DefaultTokens, Is.EqualTo(400_000));
             Assert.That(codex.Tokenizer.Id, Is.EqualTo("o200k_base"));
+            Assert.That(codex.SystemPromptRole, Is.EqualTo(SystemPromptRole.DEVELOPER));
+        });
+    }
+
+    [Test]
+    public void AModifierWhichSaysNothingAboutTheSystemPromptRoleLeavesItAlone()
+    {
+        var index = ModelFamilyIndex.Build(new SampleFamily().Rules.Concat(new FamilyWithAModifier().Rules));
+        var adjusted = index.Resolve(new ModelId("gpt-5.1-base"), LLMProviders.OPEN_AI, ModelVendor.OPEN_AI);
+
+        Assert.Multiple(() =>
+        {
+            Assert.That(adjusted.Has(Capability.FUNCTION_CALLING), Is.False, "The modifier did apply.");
+            Assert.That(adjusted.SystemPromptRole, Is.EqualTo(SystemPromptRole.DEVELOPER));
         });
     }
 
@@ -184,7 +198,8 @@ public sealed class ModelFamilyTests
                 .Apis(Capability.RESPONSES_API | Capability.CHAT_COMPLETION_API)
                 .Reasoning(ReasoningSupport.OPTIONAL)
                 .ContextWindow(400_000)
-                .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base");
+                .Tokenizer(TokenizerKind.TIKTOKEN, "o200k_base")
+                .SystemPromptRole(SystemPromptRole.DEVELOPER);
 
             builder.Rule("gpt-5.1-codex").AsPrefix().Inherits().Removes(Capability.WEB_SEARCH);
         }

app/Tests/Models/ModelProfileTests.cs

@@ -19,6 +19,7 @@ public sealed class ModelProfileTests
             Assert.That(untouched.Capabilities, Is.EqualTo(Capability.NONE));
             Assert.That(untouched.Reasoning, Is.EqualTo(ReasoningSupport.NONE));
             Assert.That(untouched.Context.IsKnown, Is.False);
+            Assert.That(untouched.SystemPromptRole, Is.EqualTo(SystemPromptRole.UNKNOWN), "The provider decides, rather than a role nobody stated.");
 
             //
             // A model we fail to recognize has to stay visible to the user rather than disappear
@@ -63,6 +64,7 @@ public sealed class ModelProfileTests
             Capabilities = Capability.TEXT_INPUT | Capability.WEB_SEARCH,
             Reasoning = ReasoningSupport.OPTIONAL,
             Context = ContextWindow.Of(128_000),
+            SystemPromptRole = SystemPromptRole.DEVELOPER,
         };
 
         var after = new ModelProfileChange { Removes = Capability.WEB_SEARCH }.ApplyTo(before);
@@ -72,6 +74,7 @@ public sealed class ModelProfileTests
             Assert.That(after.Capabilities, Is.EqualTo(Capability.TEXT_INPUT));
             Assert.That(after.Reasoning, Is.EqualTo(ReasoningSupport.OPTIONAL), "A change saying nothing about reasoning must not reset it.");
             Assert.That(after.Context, Is.EqualTo(before.Context), "A change saying nothing about the context window must not reset it.");
+            Assert.That(after.SystemPromptRole, Is.EqualTo(SystemPromptRole.DEVELOPER), "A change saying nothing about the system prompt role must not reset it.");
         });
     }
 

app/Tests/Models/OpenAI/Gpt6FamilyTests.cs

@@ -0,0 +1,50 @@
+using AIStudio.Models;
+using AIStudio.Provider;
+using AIStudio.Settings;
+// ReSharper disable InconsistentNaming
+
+namespace AIStudio.Tests.Models.OpenAI;
+
+/// <summary>
+/// Checks what AI Studio knows about the GPT-6 models when it reaches them at OpenAI itself.
+/// </summary>
+/// <remarks>
+/// OpenAI serves three GPT-6 models, and for a while the rules knew one of them. The other two fell
+/// through to the global assumption, which puts a model on the chat completion API with neither
+/// images nor web search nor reasoning. Nothing looked wrong from the outside until a request
+/// carrying a Responses API parameter was turned down.
+///
+/// The questions go through the same chain the OpenAI provider asks, assumption included, because
+/// that is where a model which no rule knows ends up looking like one somebody described.
+/// </remarks>
+[TestFixture]
+public sealed class Gpt6FamilyTests
+{
+    [TestCase("gpt-6-astra", TestName = "Astra is reached through the Responses API")]
+    [TestCase("gpt-6-sol", TestName = "Sol is reached through the Responses API")]
+    [TestCase("gpt-6-luna", TestName = "Luna is reached through the Responses API")]
+    public void EveryGpt6ModelIsAnsweredLikeTheModelPageDescribesIt(string modelId)
+    {
+        var profile = LLMProviders.OPEN_AI.GetModelProfile(new Model(modelId, null));
+
+        Assert.Multiple(() =>
+        {
+            Assert.That(profile.Has(Capability.RESPONSES_API), Is.True);
+            Assert.That(profile.Has(Capability.MULTIPLE_IMAGE_INPUT), Is.True);
+            Assert.That(profile.Has(Capability.FUNCTION_CALLING), Is.True);
+            Assert.That(profile.Has(Capability.WEB_SEARCH), Is.True);
+            Assert.That(profile.Context.IsKnown, Is.True);
+            Assert.That(profile.Context.DefaultTokens, Is.EqualTo(1_050_000));
+        });
+    }
+
+    [TestCase("gpt-6-astra", ReasoningSupport.ALWAYS, TestName = "Astra reasons on every request")]
+    [TestCase("gpt-6-sol", ReasoningSupport.ON_BY_DEFAULT, TestName = "Sol reasons unless the effort is set to none")]
+    [TestCase("gpt-6-luna", ReasoningSupport.ON_BY_DEFAULT, TestName = "Luna reasons unless the effort is set to none")]
+    public void EveryGpt6ModelReasonsTheWayItsModelPageSays(string modelId, ReasoningSupport reasoning)
+    {
+        var profile = LLMProviders.OPEN_AI.GetModelProfile(new Model(modelId, null));
+
+        Assert.That(profile.Reasoning, Is.EqualTo(reasoning));
+    }
+}

app/Tests/Models/OpenAI/NewOpenAIModelTests.cs

@@ -0,0 +1,69 @@
+using AIStudio.Models;
+using AIStudio.Provider;
+using AIStudio.Settings;
+// ReSharper disable InconsistentNaming
+
+namespace AIStudio.Tests.Models.OpenAI;
+
+/// <summary>
+/// Checks what AI Studio assumes about an OpenAI model released after its rules were written.
+/// </summary>
+/// <remarks>
+/// OpenAI releases faster than anybody writes rules, and a new generation is always reached by
+/// somebody first. What that person must get is what every current OpenAI model does: it calls
+/// tools, it looks at images, it reasons, and it answers through the Responses API. The global
+/// assumption says none of that except the tools, and it sends the model to the chat completion
+/// API, where a parameter meant for the Responses API is turned down.
+///
+/// The names below are invented on purpose. A real name belongs in the corpus; these stand for the
+/// one nobody has seen yet, which is exactly the name no corpus can hold.
+/// </remarks>
+[TestFixture]
+public sealed class NewOpenAIModelTests
+{
+    [TestCase("gpt-99", TestName = "A generation nobody wrote a rule for")]
+    [TestCase("gpt-99-mini", TestName = "A variant of a generation nobody wrote a rule for")]
+    public void ANewOpenAIModelIsAnsweredLikeTheCurrentOnes(string modelId)
+    {
+        var profile = LLMProviders.OPEN_AI.GetModelProfile(new Model(modelId, null));
+
+        Assert.Multiple(() =>
+        {
+            Assert.That(profile.Has(Capability.RESPONSES_API), Is.True);
+            Assert.That(profile.Has(Capability.FUNCTION_CALLING), Is.True);
+            Assert.That(profile.Has(Capability.MULTIPLE_IMAGE_INPUT), Is.True);
+            Assert.That(profile.Reasoning, Is.Not.EqualTo(ReasoningSupport.NONE));
+        });
+    }
+
+    [Test]
+    public void TheOpenWeightsKeepTheirOwnRuleEvenAtOpenAI()
+    {
+        //
+        // How a pattern matches is weighed before how long it is. Written as a prefix, the
+        // baseline would outrank the name part "gpt-oss" and hand the open weights reasoning and
+        // the Responses API. OpenAI does not serve them today; the test is about which rule wins.
+        //
+        var profile = LLMProviders.OPEN_AI.GetModelProfile(new Model("gpt-oss-120b", null));
+
+        Assert.Multiple(() =>
+        {
+            Assert.That(profile.Has(Capability.RESPONSES_API), Is.False);
+            Assert.That(profile.Reasoning, Is.EqualTo(ReasoningSupport.NONE));
+            Assert.That(profile.Context.DefaultTokens, Is.EqualTo(131_072));
+        });
+    }
+
+    [Test]
+    public void AModelNamedGptElsewhereIsNotTakenForAnOpenAIModel()
+    {
+        var profile = LLMProviders.SELF_HOSTED.GetModelProfile(new Model("gpt-j-6b", null));
+
+        Assert.Multiple(() =>
+        {
+            Assert.That(profile.Has(Capability.RESPONSES_API), Is.False);
+            Assert.That(profile.Has(Capability.MULTIPLE_IMAGE_INPUT), Is.False);
+            Assert.That(profile.Reasoning, Is.EqualTo(ReasoningSupport.NONE));
+        });
+    }
+}

app/Tests/Provider/OpenAI/SystemPromptRoleTests.cs

@@ -0,0 +1,48 @@
+using AIStudio.Models;
+using AIStudio.Provider;
+using AIStudio.Provider.OpenAI;
+using AIStudio.Settings;
+
+namespace AIStudio.Tests.Provider.OpenAI;
+
+/// <summary>
+/// Checks which role the OpenAI provider sends the system prompt in, model by model.
+/// </summary>
+/// <remarks>
+/// The role used to be guessed from the model's name: whatever began with an "o" or with "gpt-5",
+/// or carried a "4o", got "developer", and everything else got "system". GPT-6 fell through that
+/// guess and was sent the oldest role. The families state the role now, and these cases ask for it
+/// through the same chain the provider uses, so that a family forgetting it shows up here.
+/// </remarks>
+[TestFixture]
+public sealed class SystemPromptRoleTests
+{
+    [TestCase("gpt-6-luna", "developer")]
+    [TestCase("gpt-6-astra", "developer")]
+    [TestCase("gpt-5.6", "developer")]
+    [TestCase("gpt-4o", "developer")]
+    [TestCase("o3", "developer")]
+    [TestCase("gpt-4.1", "developer", Description = "No generation rule knows it, so the baseline for new OpenAI models answers.")]
+    [TestCase("o1-mini", "user", Description = "An early reasoning model, which takes no system prompt of its own.")]
+    [TestCase("o1-mini-2024-09-12", "user")]
+    [TestCase("o1-preview", "user")]
+    [TestCase("gpt-4", "system")]
+    [TestCase("gpt-4-turbo", "system")]
+    [TestCase("gpt-3.5-turbo", "system")]
+    public void TheSystemPromptGoesInTheRoleTheModelTakes(string modelId, string role)
+    {
+        var profile = LLMProviders.OPEN_AI.GetModelProfile(new Model(modelId, null));
+
+        Assert.That(profile.SystemPromptRole.ToOpenAIRole(), Is.EqualTo(role));
+    }
+
+    [Test]
+    public void ARoleNoRuleStatesIsTheOneOpenAIDocumentsToday()
+    {
+        //
+        // A model the rules do not know is almost always a new one, and the older generations which
+        // still want "system" say so in their families.
+        //
+        Assert.That(SystemPromptRole.UNKNOWN.ToOpenAIRole(), Is.EqualTo("developer"));
+    }
+}

documentation/Models.md

@@ -13,7 +13,7 @@ The first form is the one almost every caller wants because it includes what the
 
 ## What A Profile Says
 
-`ModelProfile` carries six things: the capabilities as a `[Flags]` enum, how the model reasons, what kind of model it is, its context window, its tokenizer, and its image limits.
+`ModelProfile` carries seven things: the capabilities as a `[Flags]` enum, how the model reasons, what kind of model it is, its context window, its tokenizer, its image limits, and the role its system prompt is sent in. Only the rules of OpenAI's own models state that role — `developer`, `system`, or, for the earliest reasoning models, a message of the person. For every other model it stays `UNKNOWN`, and the provider sends the system prompt the one way it knows.
 
 **No number ever means "unknown" by being zero.** `ContextWindow`, `TokenizerRef`, and `ImageLimits` each say so themselves — `IsKnown`, or a `null` in a nullable field. A window of zero tokens is not a thing, but zero images per message is: that is what a vLLM says before anybody raises `--limit-mm-per-prompt`. Read `ModelFactsTests` for what each of them promises.
 
@@ -74,7 +74,7 @@ public sealed class AcmeFamily : ModelFamily
 
 The source is abstract, so the compiler asks for it. That is deliberate: a rule without a page behind it is a guess, and a guess nobody can check ages into a defect. Where one page is not enough — capabilities here, the context window there, the tokenizer somewhere else — state the rest in `FurtherSources`; they are held to the same standard.
 
-What a rule can state: `Capabilities`, `Apis`, `Removes`, `Reasoning`, `Kind`, `ContextWindow`, `WithoutContextWindow`, `Tokenizer`, `Images`. What it matches: `AsExact`, `AsPrefix`, `AsSegment`, `AsSubstring`, `AlsoContains`, `NotContains`, `OnlyOn`, `OnlyFrom`.
+What a rule can state: `Capabilities`, `Apis`, `Removes`, `Reasoning`, `Kind`, `ContextWindow`, `WithoutContextWindow`, `Tokenizer`, `Images`, `SystemPromptRole`. What it matches: `AsExact`, `AsPrefix`, `AsSegment`, `AsSubstring`, `AlsoContains`, `NotContains`, `OnlyOn`, `OnlyFrom`.
 
 **Everything left unsaid stays unsaid.** A rule that says nothing about the context window does not claim that nobody knows it; it makes no statement, and whatever else does keeps its answer. `Inherits()` continues from the rule above, `InheritsFrom("<pattern>")` from a named one — worth reaching for as soon as a family has more than one generation, because "the rule above" changes when somebody inserts one.