MindWorkAI/AI-Studio
1
0 Fork 0
mirror of https://github.com/MindWorkAI/AI-Studio.git synced 2026-10-04 17:09:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Updated model capabilities across all providers (#940)

Browse Source
This commit is contained in:
Thorsten Sommer authored and GitHub committed 2026-08-29 16:07:00 +02:00
1 parent c9b7f224cc
commit edbb75dd9f
10 files changed
+275 -256
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

app/MindWork AI Studio/Settings/ProviderExtensions.Alibaba.cs
+37
View File
@@ -47,6 +47,43 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Check for the Qwen 3.8 family:
if(modelName.StartsWith("qwen3.8"))
{
// Flash thinks by default, but thinking can be turned off:
if(modelName.StartsWith("qwen3.8-flash"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT,
Capability.REASONING_BY_DEFAULT, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Unlike the open-weight checkpoint, the Max model keeps its vision
// capabilities when used through Alibaba Cloud:
if(modelName.StartsWith("qwen3.8-max"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// All other 3.8 models, such as the 27B one:
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.REASONING_BY_DEFAULT, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
// Check for the 3.0 VL models:
if(modelName.IndexOf("-vl-") is not -1)
return
app/MindWork AI Studio/Settings/ProviderExtensions.Anthropic.cs
+23 -2
View File
@@ -18,6 +18,26 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Claude Opus 5 and Sonnet 5 think adaptively unless thinking is turned off:
if(modelName.StartsWith("claude-opus-5") || modelName.StartsWith("claude-sonnet-5"))
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.REASONING_BY_DEFAULT, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Claude Haiku 4.5 needs an explicit thinking budget to reason:
if(modelName.StartsWith("claude-haiku-4-5"))
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Claude 4.x models:
if(modelName.StartsWith("claude-opus-4") || modelName.StartsWith("claude-sonnet-4"))
return [
@@ -48,9 +68,10 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Any other model is able to process text only:
// Any other model. Every current Claude model accepts images, so we assume the
// same for models we do not know yet:
return [
Capability.TEXT_INPUT,
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
app/MindWork AI Studio/Settings/ProviderExtensions.DeepSeek.cs
+19 -9
View File
@@ -8,21 +8,31 @@ public static partial class ProviderExtensions
{
var modelName = model.Id.ToLowerInvariant().AsSpan();
// The reasoner alias points to the thinking mode of the current flash model:
if(modelName.IndexOf("reasoner") is not -1)
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
// The chat alias points to the non-thinking mode of the same model:
if(modelName.IndexOf("chat") is not -1)
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// DeepSeek publishes its models as open weights and offers them under the same
// names here. Instead of maintaining a second copy of those rules, we reuse the
// ones for open source models:
return GetModelCapabilitiesOpenSource(model);
}
}
app/MindWork AI Studio/Settings/ProviderExtensions.Google.cs
+6 -7
View File
@@ -10,14 +10,13 @@ public static partial class ProviderExtensions
if (modelName.IndexOf("gemini-") is not -1)
{
// Chat-compatible Gemini 3.x reasoning models:
if (modelName is "gemini-3.5-flash" ||
// Chat-compatible Gemini 3.x reasoning models. We match the entire 3.x line
// so that new releases are covered as well: they all reason, and the
// thinking level can only be lowered, never turned off. The two rolling
// aliases carry no version number and are listed separately:
if (modelName.IndexOf("gemini-3") is not -1 ||
modelName is "gemini-flash-latest" ||
modelName is "gemini-3.1-flash-lite" ||
modelName is "gemini-3-flash-preview" ||
modelName is "gemini-pro-latest" ||
modelName is "gemini-3.1-pro-preview" ||
modelName is "gemini-3.1-pro-preview-customtools")
modelName is "gemini-pro-latest")
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.AUDIO_INPUT,
app/MindWork AI Studio/Settings/ProviderExtensions.OpenAI.cs
+2 -2
View File
@@ -182,7 +182,7 @@ public static partial class ProviderExtensions
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.FUNCTION_CALLING, Capability.OPTIONAL_REASONING, Capability.REASONING_BY_DEFAULT,
Capability.FUNCTION_CALLING, Capability.REASONING_BY_DEFAULT,
Capability.WEB_SEARCH,
Capability.RESPONSES_API, Capability.CHAT_COMPLETION_API,
];
@@ -193,7 +193,7 @@ public static partial class ProviderExtensions
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.OPTIONAL_REASONING, Capability.REASONING_BY_DEFAULT,
Capability.FUNCTION_CALLING, Capability.REASONING_BY_DEFAULT,
Capability.WEB_SEARCH,
Capability.RESPONSES_API, Capability.CHAT_COMPLETION_API,
];
app/MindWork AI Studio/Settings/ProviderExtensions.OpenRouter.cs
+55 -233
View File
@@ -6,244 +6,66 @@ public static partial class ProviderExtensions
{
private static List<Capability> GetModelCapabilitiesOpenRouter(Model model)
{
var modelName = model.Id.ToLowerInvariant().AsSpan();
//
// OpenRouter model IDs follow the pattern "vendor/model-name". Examples:
// - openai/gpt-5.6
// - anthropic/claude-opus-5
// - google/gemini-3.7-flash
// - qwen/qwen3.8-flash-next
//
// OpenRouter offers the models of all the other providers. Instead of keeping a
// second set of rules here, which would always lag behind, we hand the model
// over to the provider implementation which already knows it. The vendor prefix
// has to be removed first: some of those implementations match the beginning of
// the model name and would not recognize a prefixed ID.
//
var separatorIndex = model.Id.IndexOf('/');
var vendor = separatorIndex is -1 ? string.Empty : model.Id[..separatorIndex].ToLowerInvariant();
var bareModel = separatorIndex is -1 ? model : model with { Id = model.Id[(separatorIndex + 1)..] };
var bareModelName = bareModel.Id.ToLowerInvariant().AsSpan();
//
// OpenRouter model IDs follow the pattern: "provider/model-name"
// Examples:
// - openai/gpt-4o
// - anthropic/claude-3-5-sonnet
// - google/gemini-pro-1.5
// - meta-llama/llama-3.1-405b-instruct
//
// We need to detect capabilities based on both provider and model name.
//
//
// OpenAI models via OpenRouter:
//
if (modelName.IndexOf("openai/") is not -1)
var capabilities = vendor switch
{
// Reasoning models (o1, o3, o4 series)
if (modelName.IndexOf("/o1") is not -1 ||
modelName.IndexOf("/o3") is not -1 ||
modelName.IndexOf("/o4") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.CHAT_COMPLETION_API,
];
// The gpt-oss models are open weights. The OpenAI implementation does not
// know them, because they are not part of the OpenAI cloud offering:
"openai" when bareModelName.IndexOf("gpt-oss") is not -1 => GetModelCapabilitiesOpenSource(bareModel),
"openai" => GetModelCapabilitiesOpenAI(bareModel),
// GPT-4o and GPT-5 series with multimodal
if (modelName.IndexOf("/gpt-4o") is not -1 ||
modelName.IndexOf("/gpt-5") is not -1 ||
modelName.IndexOf("/chatgpt-4o") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
"anthropic" => GetModelCapabilitiesAnthropic(bareModel),
// Standard GPT-4
if (modelName.IndexOf("/gpt-4") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Gemma is open weights, Gemini is not:
"google" when bareModelName.IndexOf("gemma") is not -1 => GetModelCapabilitiesOpenSource(bareModel),
"google" => GetModelCapabilitiesGoogle(bareModel),
// GPT-3.5
if (modelName.IndexOf("/gpt-3.5") is not -1 ||
modelName.IndexOf("/gpt-3") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
"mistralai" => GetModelCapabilitiesMistral(bareModel),
"perplexity" => GetModelCapabilitiesPerplexity(bareModel),
//
// Anthropic models via OpenRouter:
//
if (modelName.IndexOf("anthropic/") is not -1)
{
// Claude 3.5 and newer with vision
if (modelName.IndexOf("/claude-3.5") is not -1 ||
modelName.IndexOf("/claude-3-5") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Everything else is open source: Qwen, Llama, GLM, Kimi, Muse, Hunyuan,
// Nemotron, Grok, and whatever OpenRouter adds next. DeepSeek belongs here
// as well: its own implementation covers the aliases of the DeepSeek
// platform, while OpenRouter uses the names of the open weights.
_ => GetModelCapabilitiesOpenSource(bareModel),
};
// Claude 3 Opus/Sonnet with vision
if (modelName.IndexOf("/claude-3-opus") is not -1 ||
modelName.IndexOf("/claude-3-sonnet") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Other Claude 3 models
if (modelName.IndexOf("/claude-3") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
//
// Google models via OpenRouter:
//
if (modelName.IndexOf("google/") is not -1)
{
// Gemini models with multimodal
if (modelName.IndexOf("/gemini") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
//
// xAI Grok models via OpenRouter:
//
if (modelName.IndexOf("x-ai/") is not -1 || modelName.IndexOf("/grok") is not -1)
{
if (modelName.IndexOf("-vision") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
//
// DeepSeek models via OpenRouter:
//
if (modelName.IndexOf("/deepseek") is not -1)
{
if (modelName.IndexOf("-r1") is not -1 || modelName.IndexOf(" r1") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.CHAT_COMPLETION_API,
];
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
//
// Mistral models via OpenRouter:
//
if (modelName.IndexOf("/mistral") is not -1 || modelName.IndexOf("/pixtral") is not -1)
{
if (modelName.IndexOf("/pixtral") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
//
// Meta Llama models via OpenRouter:
//
if (modelName.IndexOf("/llama") is not -1)
{
// Llama 4 with vision
if (modelName.IndexOf("/llama-4") is not -1 ||
modelName.IndexOf("/llama4") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Vision models
if (modelName.IndexOf("-vision") is not -1)
return [
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
// Llama 3.1+ with function calling
if (modelName.IndexOf("/llama-3.") is not -1 ||
modelName.IndexOf("/llama3.") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Default Llama
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
//
// Qwen models via OpenRouter:
//
if (modelName.IndexOf("/qwen") is not -1 || modelName.IndexOf("/qwq") is not -1)
{
if (modelName.IndexOf("/qwq") is not -1)
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.CHAT_COMPLETION_API,
];
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
//
// Default for unknown models:
// Assume basic text input/output with chat completion
//
return [
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
return NormalizeForOpenRouter(capabilities);
}
}
/// <summary>
/// Adjusts the capabilities reported by another provider for use through OpenRouter.
/// </summary>
/// <param name="capabilities">The capabilities as reported by the provider implementation.</param>
/// <returns>The capabilities as they apply when using the model through OpenRouter.</returns>
/// <remarks>
/// OpenRouter serves every model through its OpenAI-compatible chat completion API.
/// The Responses API is not available there, no matter which API the original
/// provider offers.
/// </remarks>
private static List<Capability> NormalizeForOpenRouter(List<Capability> capabilities)
{
capabilities.Remove(Capability.RESPONSES_API);
if(!capabilities.Contains(Capability.CHAT_COMPLETION_API))
capabilities.Add(Capability.CHAT_COMPLETION_API);
return capabilities;
}
}
app/MindWork AI Studio/Settings/ProviderExtensions.OpenSource.cs
+100 -2
View File
@@ -71,6 +71,21 @@ public static partial class ProviderExtensions
];
}
//
// Meta Muse models. They need their own block because their names do not
// contain "llama". Muse Glimmer always reasons: its chat template opens the
// thinking channel unconditionally, only the reasoning strength can be lowered.
//
if (modelName.IndexOf("muse-glimmer") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
//
// DeepSeek models:
//
@@ -122,6 +137,19 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Check for the Qwen 3.8 Flash models. The open weights are published as
// Flash-Next, while Flash without the suffix is the production model. Both
// share the same capabilities, so one check covers them:
if(modelName.IndexOf("qwen3.8-flash") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT,
Capability.REASONING_BY_DEFAULT, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Check for the multimodal Qwen 3.8 27B checkpoint:
if(modelName.IndexOf("qwen3.8-27b") is not -1)
return
@@ -193,6 +221,23 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
//
// Tencent Hunyuan models. Hy3 answers directly by default: its reasoning_effort
// parameter defaults to no_think, low and high must be requested. We also match
// the short name because providers offer the model as tencent/hy3, so checking
// the start of the name is not enough.
//
if (modelName.IndexOf("hunyuan") is not -1 ||
modelName.IndexOf("hy3") is not -1)
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
//
// Mistral models:
//
@@ -335,6 +380,18 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Grok 4 models take text, images, and video natively. Reasoning is always
// on, only the reasoning effort can be configured:
if(modelName.IndexOf("grok-4") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
if(modelName.StartsWith("grok-3-mini"))
return
[
@@ -346,14 +403,25 @@ public static partial class ProviderExtensions
];
if(modelName.StartsWith("grok-3"))
return
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Any other Grok model. Without this, unknown Grok versions would fall
// through to the global default and would lose function calling:
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
//
@@ -383,11 +451,41 @@ public static partial class ProviderExtensions
];
}
//
// NVIDIA Nemotron models. They are built for agentic workloads and are text
// only. Reasoning has to be requested through enable_thinking, so it is
// optional. The check also covers the quantized checkpoints such as
// NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4.
//
if (modelName.IndexOf("nemotron") is not -1)
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.OPTIONAL_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
//
// Z AI / GLM models:
//
if (modelName.IndexOf("glm") is not -1)
{
// GLM 5.3 uses forced thinking: the reasoning effort can be lowered, but
// reasoning cannot be turned off. This check must stay in front of the
// vision check below, because quantized builds such as GLM-5.3-Flash-NVFP4
// contain a "v" and would be misread as a vision model:
if (modelName.IndexOf("glm-5.3") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
if (modelName.IndexOf("glm-5.2") is not -1)
return
[