Keep non-chat models out of the chat lists

This commit is contained in:
Thorsten Sommer committed 2026-09-12 13:52:10 +02:00
1 parent 77130a450a
commit b6defc8b4b
11 files changed
+111 -5

No files matched your search

@@ -168,5 +168,14 @@ public static class ExpectedChanges
AnswerWanted: [TEXT_INPUT, MULTIPLE_IMAGE_INPUT, TEXT_OUTPUT, FUNCTION_CALLING, WEB_SEARCH, RESPONSES_API],
Reason: "The alias for the non-reasoning GPT-5 is claimed by the \"gpt-5-\" prefix rule and is told it always reasons, which is the one thing its name rules out.",
Source: "OpenAI names this alias as the non-reasoning model of the GPT-5 line; the corpus entry \"gpt-5\" next to it is the reasoning one."),
//
// A model nobody had written a rule for yet, found while testing the switch-over.
//
new(X, "grok-build-0.1",
AnswerToday: [TEXT_INPUT, TEXT_OUTPUT, FUNCTION_CALLING, CHAT_COMPLETION_API],
AnswerWanted: [TEXT_INPUT, MULTIPLE_IMAGE_INPUT, TEXT_OUTPUT, FUNCTION_CALLING, CHAT_COMPLETION_API],
Reason: "The agentic coding model of the Grok line reads pictures, and the family fallback it reaches says text only.",
Source: "https://x.ai/news/grok-build-0-1 states text and image input, tool calling, and a 256K context window."),
];
}
+1
View File
@@ -204,6 +204,7 @@ public static class ModelCorpus
new(X, "grok-3-mini", NAMED_BY_A_RULE),
new(X, "grok-2-vision-1212", NAMED_BY_A_RULE),
new(X, "grok-5", NAMED_BY_NO_RULE),
new(X, "grok-build-0.1", NAMED_BY_A_RULE),
];
/// <summary>
@@ -69,6 +69,9 @@ public static class ModelKindCorpus
new(OPEN_AI, "sora-2", VIDEO_GENERATION),
new(GOOGLE, "veo-3.0-generate-001", VIDEO_GENERATION),
new(SELF_HOSTED, "kling-video-v2", VIDEO_GENERATION),
new(X, "grok-imagine-video", VIDEO_GENERATION, AnsweredTodayAs: CHAT, Reason: "Found in the chat list while testing. No marker knew the name, and the xAI provider only kept models whose name lacks \"-image\" -- which \"-imagine\" does."),
new(X, "grok-imagine-video-1.5", VIDEO_GENERATION, AnsweredTodayAs: CHAT, Reason: "The same, one version on."),
];
/// <summary>
@@ -108,6 +111,16 @@ public static class ModelKindCorpus
// The name the marker file names as the reason for asking this question before the others:
new(OPEN_AI, "gpt-realtime-whisper", REALTIME),
new(OPEN_AI, "gpt-live-1", REALTIME, AnsweredTodayAs: CHAT, Reason: "Found in the chat list while testing. The line which succeeds the realtime models dropped the word, and it is even less of a chat partner: it listens and speaks at once and leaves the thinking to a text model behind it."),
];
/// <summary>
/// The models which work a screen.
/// </summary>
private static readonly ModelKindExample[] COMPUTER_USE_ENTRIES =
[
new(GOOGLE, "gemini-2.5-computer-use-preview-10-2025", COMPUTER_USE, AnsweredTodayAs: CHAT, Reason: "Found in the chat list while testing. Its API refuses every request which does not carry the computer use tool, so a conversation with it cannot even begin."),
];
/// <summary>
@@ -161,6 +174,19 @@ public static class ModelKindCorpus
new(SELF_HOSTED, "llama-2-7b-chat-klingon", CHAT),
new(SELF_HOSTED, "llama3.3:70b", CHAT),
new(OPEN_AI, "gpt-5.1", CHAT),
//
// Three which were questioned while testing and stay all the same. Grok Build is the coding
// model behind the xAI CLI and answers like any other Grok. The Groq compound systems are
// models with tools already built in, reached through the ordinary chat completion API. And
// Gemini Robotics ER answers in text; it is built for pointing at things in a picture rather
// than for conversation, but a conversation with it works, and a model which works belongs
// in the list.
//
new(X, "grok-build-0.1", CHAT),
new(GROQ, "groq/compound", CHAT),
new(GROQ, "groq/compound-mini", CHAT),
new(GOOGLE, "gemini-robotics-er-1.5-preview", CHAT),
];
/// <summary>
@@ -175,6 +201,7 @@ public static class ModelKindCorpus
..TRANSCRIPTION_ENTRIES,
..SPEECH_ENTRIES,
..REALTIME_ENTRIES,
..COMPUTER_USE_ENTRIES,
..TEXT_COMPLETION_ENTRIES,
..OCR_ENTRIES,
..MODERATION_ENTRIES,