chore: sync new models#918
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Codex (@codex) review |
![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: 2a406a57c4
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 (@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 (@codex) address that feedback".
| "displayName": "GPT-5.4 mini", | ||
| "reasoning": true, | ||
| "max_input_tokens": 400000, | ||
| "max_input_tokens": 272000, |
There was a problem hiding this comment.
OpenAI's GPT-5.4 mini and nano docs list a 400,000-token context window with a separate 128,000 max-output limit (https://developers.openai.com/api/docs/models/gpt-5.4-mini, https://developers.openai.com/api/docs/models/gpt-5.4-nano), and this catalog stores the context window in max_input_tokens for sibling models like gpt-5.4/gpt-5.5 rather than subtracting output tokens. Setting the aliases to 272,000 under-advertises the supported input window, so consumers of the published schema can reject or avoid valid prompts between 272k and 400k tokens; the same 272k replacement should be reverted for the snapshot and nano alias changed in this commit.
Useful? React with 👍 / 👎.
| ], | ||
| "max_input_tokens": 1000000, | ||
| "max_output_tokens": 128000, | ||
| "max_output_tokens": 64000, |
There was a problem hiding this comment.
Anthropic documents Claude Sonnet 4.6 as supporting up to 128k output tokens (https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking), and the Bedrock/Vertex aliases in this same catalog still advertise 128000. Lowering only the direct Anthropic entry to 64000 makes schema consumers think 128k is unsupported and also causes openaiParamsToAnthropicMesssageParams to default omitted max_tokens requests to 64k for this model, truncating workloads that relied on the documented 128k limit.
Useful? React with 👍 / 👎.
| "input_cost_per_mil_tokens": 1.4, | ||
| "output_cost_per_mil_tokens": 4.4, | ||
| "input_cache_read_cost_per_mil_tokens": 0.14, | ||
| "input_cache_read_cost_per_mil_tokens": 0.26, |
There was a problem hiding this comment.
This entry is the Fireworks-specific accounts/fireworks/models/glm-5p2 model, but Fireworks' pricing page lists GLM 5.2 Standard at $1.40 input / $0.14 cached input / $4.40 output per 1M tokens (https://docs.fireworks.ai/serverless/pricing). Changing cached-input pricing to 0.26 overstates cache-read cost for Fireworks traffic and makes downstream cost estimates or billing displays wrong for cached prompts.
Useful? React with 👍 / 👎.
Codex Review Response for PR #918All three Codex findings were valid. The automated sync introduced five regressions that have been reverted in Changes Made1. Restored GPT-5.4 mini/nano
|
github-actions
Bot
force-pushed
the
chore/sync-new-models
branch
from
2a406a5 to
3b044f4
Compare
|
Codex Review: Didn't find any major issues. 🎉 Reviewed commit: ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
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 (@codex) address that feedback". |
Automated sync of model metadata and newly available models from the approved provider allowlist.