feat(llm): add GPT-5.6 model support - #4
Conversation
![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: 4db7c82e4d
ℹ️ About Codex in GitHub
Your team has set up Codex to 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 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| questionary.Choice("Low (Faster)", "low"), | ||
| questionary.Choice("None (No reasoning)", "none"), | ||
| questionary.Choice("Extra high (More thorough)", "xhigh"), | ||
| questionary.Choice("Max (Most thorough)", "max"), |
There was a problem hiding this comment.
Selecting Max here is not valid for every OpenAI model still offered by this same flow: the catalog retains pre-5.6 choices such as gpt-5.5 and gpt-5.4, while OpenAI's model pages list those models as supporting none, low, medium, high, and xhigh but not max (for example, https://developers.openai.com/api/docs/models/gpt-5.5). Because the selected openai_reasoning_effort is forwarded unchanged to both OpenAI clients, choosing Max with any retained older quick/deep model will make the run fail at the API request; restrict this option to GPT-5.6 models or normalize it before constructing the clients.
Useful? React with 👍 / 👎.
1 participant
Summary
Impact
New runs use the role-appropriate GPT-5.6 variants by default across the library, runner, backend, web frontend, examples, and structured-output smoke script. Existing configurations that name older models continue to work.
Validation
.venv/bin/pytest -q(693 passed,2 skipped)56 passed,69 subtests passed).venv/bin/ruff check ...npm run buildinweb/frontendgit diff --checkNo live OpenAI API call was performed.