feat: sync 10 missing models + fix watermark - #63
Conversation
- Add 10 missing models released between 2026-06-30 and 2026-07-15 - OpenAI: GPT-5.6 Sol/Terra/Luna (+ Pro variants) — 6 models - Anthropic: Claude Sonnet 5 - Google: Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite) - xAI: Grok 4.5 - Tencent: Hy3 (stable release, distinct from hy3-preview) - Update .sync-watermark from 1782276303 to 1784554658 - presetDataVersion: 79 → 80
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 768da94ad0
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| "match": { "exact": ["gemini-3.1-flash-lite-image"], "patterns": ["gemini-3.1-flash-lite-image*"] }, | ||
| "spec": { | ||
| "contextWindow": 65536, | ||
| "maxOutputTokens": 65536, |
There was a problem hiding this comment.
Google's Gemini API model page for gemini-3.1-flash-lite-image (https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image) lists the output token limit as 4,096, so advertising 65,536 here lets callers size max_tokens above the API limit. When clients rely on this spec for request limits, long generations can be rejected by the Google/Gemini backend instead of being capped correctly.
Useful? React with 👍 / 👎.
| "capabilities": ["chat", "reasoning", "code", "tool_use", "agent", "vision", "long_context"], | ||
| "serviceType": ["chat"], |
There was a problem hiding this comment.
When a cloud/NewAPI model named claude-sonnet-5 is enriched from model-specs, this entry overrides serviceType/capabilities without computer_use; the provider entry for the same model already declares chat plus computer_use and includes the capability. In that scenario, clients will hide or fail to select Sonnet 5 for computer-use workflows even though the direct Anthropic config supports it.
Useful? React with 👍 / 👎.
| "spec": { | ||
| "contextWindow": 65536, | ||
| "maxOutputTokens": 65536, | ||
| "capabilities": ["chat", "vision", "image_generation"], |
There was a problem hiding this comment.
Google's docs for gemini-3.1-flash-lite-image list image editing and multi-turn image editing as supported, but this spec only tags generation. When clients filter models by image_editing capability before sending an edit request, Nano Banana 2 Lite will be excluded even though it supports the workflow.
Useful? React with 👍 / 👎.
| "contextWindow": 500000, | ||
| "maxOutputTokens": 128000, | ||
| "capabilities": ["chat", "reasoning", "code", "tool_use", "vision", "long_context"], | ||
| "serviceType": ["chat"], |
There was a problem hiding this comment.
xAI's reasoning docs say grok-4.5 supports reasoning_effort and that reasoning cannot be disabled, but this serviceType omits reasoning. When cloud/NewAPI models are enriched from this spec and clients use serviceType for routing or filtering reasoning models, Grok 4.5 will be classified as plain chat and skipped despite requiring/using reasoning behavior.
Useful? React with 👍 / 👎.
1 participant
变更
修复水位线未更新导致遗漏模型的 bug，补入水位线之后所有已追踪厂商的新模型。
Bug 修复
.sync-watermark，导致后续增量扫描始终从 6月24日起算，7月发布的新模型全部被漏掉created时间戳1782276303(2026-06-24) →1784554658(2026-07-21)新增 10 个模型
其他