🤖 feat: add GPT-5.1-Codex-Max model with xhigh reasoning level #933
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
mux/src/browser/utils/commands/sources.ts
Lines 452 to 454 in ecc4440
xhigh palette option saved as off
The command palette now exposes an xhigh thinking level (mapped over THINKING_LEVELS here), but App.tsx still normalizes palette submissions against a four-value THINKING_LEVELS = ["off","low","medium","high"] when persisting (setThinkingLevelFromPalette around lines 44/300). Selecting xhigh for any model therefore gets coerced to "off" before it reaches the thinking provider, so the newly added level can’t actually be used.
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".
|
Gonna finish this off as I assume you're out for the weekend |
Add support for OpenAI's gpt-5.1-codex-max model which introduces a new 'xhigh' (Extra High) reasoning effort level for extended deep thinking. Changes: - Extended ThinkingLevel type to include 'xhigh' as 5th option - Added GPT_CODEX_MAX to known models with codex-max alias - Updated thinking policy to return 5 levels for codex-max only - Added xhigh to UI level descriptions and command palette - Added model pricing/capability data to models-extra - Updated SendMessageOptionsSchema to accept xhigh - Added comprehensive tests for codex-max policy The xhigh level is exclusive to gpt-5.1-codex-max. Other models gracefully fall back to their maximum supported level when xhigh is requested. Change-Id: Iab7ba7187703e275c4c0aa76779381dff4006316 Signed-off-by: Thomas Kosiewski <tk@coder.com>
Same pricing as gpt-5.1-codex: $1.25/M input, $10/M output Also aligned max token limits with the codex model. Change-Id: I030014df05a5ccae62f5c93d7435ec2363d23317 Signed-off-by: Thomas Kosiewski <tk@coder.com>
ammar-agent
force-pushed
the
add-gpt-5-1-codex-max-model
branch
from
20c0858 to
bc841f9
Compare
3 participants
Add support for OpenAI's
gpt-5.1-codex-maxmodel with the newxhigh(Extra High) thinking level.Changes
xhighthinking level end-to-end (types, schemas, UI policy, enforcement)gpt-5.1-codex-maxworks with/without theopenai:prefix orcodex-maxaliasmodels.json) to include codex-maxdocs/models.mdBehavior
xhighonly surfaces for codex-max; other models fall back to their allowed max (e.g., medium or high)enforceThinkingPolicyValidation
make typecheckbun test src/browser/utils/thinking/policy.test.tsbun run scripts/update_models.ts(models.json refreshed)Generated with
mux