MindWorkAI/AI-Studio
MindWorkAI/AI-Studio
1
0
Fork 0
mirror of https://github.com/MindWorkAI/AI-Studio.git synced 2026-09-27 02:53:38 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Correct capabilities of cloud provider models

Browse Source
This commit is contained in:
Thorsten Sommer 2026-09-11 10:25:42 +02:00
parent cab1dd1609
commit 5948c50697
Signed by untrusted user who does not match committer: tsommer
GPG Key ID: 371BBA77A02C0108
5 changed files with 191 additions and 54 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

72
app/MindWork AI Studio/Settings/ProviderExtensions.Alibaba.cs
View File

@ -11,8 +11,24 @@ public static partial class ProviderExtensions
// Qwen models:
if (modelName.StartsWith("qwen"))
{
// Check for omni models:
// Check for omni models. Alibaba lists the Qwen3 and Qwen3.5 Omni series among the
// models which call functions; the older qwen-omni ones are not on that list, which
// is what the version check separates here:
if (modelName.IndexOf("omni") is not -1)
{
if (modelName.StartsWith("qwen3"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.AUDIO_INPUT, Capability.SPEECH_INPUT,
Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT, Capability.SPEECH_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
@ -23,6 +39,7 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
}
// Check for Qwen 3.5:
if(modelName.StartsWith("qwen3.5"))
@ -47,6 +64,44 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Check for the Qwen 3.7 family. Thinking is optional here and switched on by
// default, except for the two preview snapshots, which do nothing else:
if(modelName.StartsWith("qwen3.7"))
{
if(modelName.IndexOf("-preview") is not -1 ||
modelName.IndexOf("-2026-05-17") is not -1)
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Vision arrived in the middle of the series. The rolling qwen3.7-max alias
// still answers as the text-only May snapshot, so only the June one may be
// told that it reads images and video:
if(modelName.IndexOf("-2026-06-08") 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,
];
return
[
Capability.TEXT_INPUT,
Capability.TEXT_OUTPUT,
Capability.REASONING_BY_DEFAULT, Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
}
// Check for the Qwen 3.8 family:
if(modelName.StartsWith("qwen3.8"))
{
@ -84,8 +139,20 @@ public static partial class ProviderExtensions
];
}
// Check for the 3.0 VL models:
// Check for the VL models. Alibaba names the Qwen3-VL Plus and Flash series as
// function callers; the older qwen-vl models are absent from that list:
if(modelName.IndexOf("-vl-") is not -1)
{
if(modelName.StartsWith("qwen3"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
@ -93,6 +160,7 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
}
// Check for Qwen 3:
if(modelName.StartsWith("qwen3"))

82
app/MindWork AI Studio/Settings/ProviderExtensions.Google.cs
View File

@ -10,10 +10,57 @@ public static partial class ProviderExtensions
if (modelName.IndexOf("gemini-") is not -1)
{
//
// Image generation models. They carry a version number like every other model and
// have to be asked about first, or gemini-3-pro-image would be read as a chat model
// of the 3.x line and be promised function calling. No image model of the family
// offers that; what they do offer, and the chat models do not, is writing images.
//
if (modelName.IndexOf("-image") is not -1)
{
// Of the image models, only the 3.1 Flash ones read video. They think about
// complex prompts, and, as with the 3.x chat models, thinking cannot be
// switched off:
if (modelName.IndexOf("gemini-3.1-flash") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.CHAT_COMPLETION_API,
];
// Every other Gemini 3 image model thinks as well, it just does not read video:
if (modelName.IndexOf("gemini-3") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.CHAT_COMPLETION_API,
];
// The older image models, such as the 2.5 Flash one, do not think:
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
}
// 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:
// thinking level can only be lowered, never turned off. That holds for the
// Flash Lite models of this line too, which is what sets them apart from
// Gemini 2.5 Flash Lite below: there, thinking is off until it is asked for,
// while here the lowest level still thinks. 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-pro-latest")
@ -54,17 +101,6 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// Image generation:
if(modelName.IndexOf("-2.0-flash-preview-image-") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.AUDIO_INPUT,
Capability.SPEECH_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
// Realtime model:
if(modelName.IndexOf("-2.0-flash-live-") is not -1)
return
@ -78,16 +114,16 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
// The 2.0 flash models cannot call functions:
if(modelName.IndexOf("-2.0-flash-") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.AUDIO_INPUT,
Capability.SPEECH_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_OUTPUT,
Capability.CHAT_COMPLETION_API,
];
//
// There used to be a branch here which withheld function calling from the 2.0 Flash
// models. It said the wrong thing about them, and it only ever caught the dated IDs
// because it asked for a trailing hyphen: the plain gemini-2.0-flash alias walked
// past it and got a different answer than gemini-2.0-flash-001, which is the same
// model. Both questions are moot now, because Google shut the 2.0 Flash chat models
// down on 1 June 2026. Anything still asking for one of those names gets the default
// below. The live model above keeps its branch: it belongs to a different API whose
// retirement Google announces separately.
//
// The old 1.0 pro vision model:
if(modelName.IndexOf("pro-vision") is not -1)

37
app/MindWork AI Studio/Settings/ProviderExtensions.OpenAI.cs
View File

@ -54,14 +54,6 @@ public static partial class ProviderExtensions
Capability.CHAT_COMPLETION_API,
];
if (modelName.StartsWith("chatgpt-4o-"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.RESPONSES_API,
];
if (modelName.StartsWith("o3-mini"))
return
[
@ -118,6 +110,7 @@ public static partial class ProviderExtensions
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.ALWAYS_REASONING,
Capability.WEB_SEARCH,
Capability.RESPONSES_API,
];
@ -132,11 +125,18 @@ public static partial class ProviderExtensions
Capability.RESPONSES_API,
];
//
// None of the GPT-5 models writes images itself. They can ask for one through the
// image generation tool, which is a tool call like any other and produces a picture
// from a separate model. That is a different thing from an output modality, and we
// must not report it as one: the chat would then offer to receive images which never
// arrive.
//
if(modelName is "gpt-5.1" || modelName.StartsWith("gpt-5.1-"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.OPTIONAL_REASONING,
Capability.WEB_SEARCH,
@ -147,7 +147,7 @@ public static partial class ProviderExtensions
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.OPTIONAL_REASONING,
Capability.WEB_SEARCH,
@ -180,7 +180,7 @@ public static partial class ProviderExtensions
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT, Capability.IMAGE_OUTPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.REASONING_BY_DEFAULT,
Capability.WEB_SEARCH,
@ -198,6 +198,21 @@ public static partial class ProviderExtensions
Capability.RESPONSES_API, Capability.CHAT_COMPLETION_API,
];
//
// GPT-6 Astra. 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.
//
if(modelName is "gpt-6-astra" || modelName.StartsWith("gpt-6-astra-"))
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING, Capability.ALWAYS_REASONING,
Capability.WEB_SEARCH,
Capability.RESPONSES_API, Capability.CHAT_COMPLETION_API,
];
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,

21
app/MindWork AI Studio/Settings/ProviderExtensions.OpenSource.cs
View File

@ -547,12 +547,27 @@ 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:
// One member of the 4.20 line answers without thinking, and it says so in its
// name. It has to be asked about before the general Grok 4 rule, which would
// otherwise claim the opposite of what the name states:
if(modelName.IndexOf("-non-reasoning") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.FUNCTION_CALLING,
Capability.CHAT_COMPLETION_API,
];
// Grok 4 models take text and images. Reasoning is always on, only the
// reasoning effort can be configured. Video is not among their modalities:
// xAI serves audio, image, and video through models and APIs of their own,
// and the model pages of the 4.x line say "text, image" and nothing else:
if(modelName.IndexOf("grok-4") is not -1)
return
[
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT, Capability.VIDEO_INPUT,
Capability.TEXT_INPUT, Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.ALWAYS_REASONING, Capability.FUNCTION_CALLING,

7
app/MindWork AI Studio/Settings/ProviderExtensions.Perplexity.cs
View File

@ -8,6 +8,11 @@ public static partial class ProviderExtensions
{
var modelName = NormalizeModelId(model.Id).AsSpan();
//
// No Sonar model writes images. What looked like it does is the option to have the
// answer come with images: those are pictures the search found on the pages it read,
// handed back as links, not something the model drew.
//
if(modelName.IndexOf("reasoning") is not -1 ||
modelName.IndexOf("deep-research") is not -1)
return
@ -16,7 +21,6 @@ public static partial class ProviderExtensions
Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.IMAGE_OUTPUT,
Capability.ALWAYS_REASONING,
Capability.WEB_SEARCH,
@ -29,7 +33,6 @@ public static partial class ProviderExtensions
Capability.MULTIPLE_IMAGE_INPUT,
Capability.TEXT_OUTPUT,
Capability.IMAGE_OUTPUT,
Capability.WEB_SEARCH,
Capability.CHAT_COMPLETION_API,