tt/ttstd-infinite-canvas
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: recognize common multimodal text models

Browse Source
This commit is contained in:
Codex
2026-09-19 20:40:55 +08:00
parent 5843d489c6
commit 10b85962b6
4 changed files with 68 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
web/test/model-capabilities.test.ts
View File

@@ -31,3 +31,16 @@ test("raising the video default leaves text and image limits untouched", () => {
assert.equal(profile.text!.references.promptMaxChars, 32000);
assert.equal(profile.image!.references.promptMaxChars, 32000);
});
test("known multimodal text model families default to image input", () => {
const gemini = defaultModelCapabilityConfig("chat-completion", "gemini-3.8-flash-high").text!;
const gpt = defaultModelCapabilityConfig("chat-completion", "gpt-4o").text!;
const doubao = defaultModelCapabilityConfig("chat-completion", "doubao-seed-1.6-flash").text!;
const plain = defaultModelCapabilityConfig("chat-completion", "plain-text-model").text!;
assert.equal(gemini.references.maxImages, 16);
assert.equal(gemini.references.maxImageBytes, 30 * 1024 * 1024);
assert.equal(gpt.references.maxImages, 16);
assert.equal(doubao.references.maxImages, 16);
assert.equal(plain.references.maxImages, 0);
});