You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: Auto-generated commits fail when user lacks access to hardcoded gpt-5.3-codex model #1165
I searched existing issues and did not find a duplicate.
I included enough detail to reproduce or investigate the problem.
Area
apps/server
Steps to reproduce
Start the T3 Code app.
Open a project with unstaged/staged Git changes.
Select a specific model in the chat UI model picker at the bottom (e.g., gpt-5.2-codex).
Click the "Auto Commit" or "Commit & Push" button in the Git actions block.
Expected behavior
The auto-generated commit message should be created using the model selected in the chat UI (gpt-5.2-codex), as that is what the user has chosen and verified they have access to.
Actual behavior
The app ignores the selected model and attempts to generate the commit message using a hardcoded gpt-5.3-codex model.
Impact
Major degradation or frequent failure
Version or commit
main
Environment
Provider/Model: Codex (gpt-5.3-codex hardcoded, user attempted gpt-5.2-codex)
Logs or stack traces
// The model is currently hardcoded in apps/server/src/git/Layers/CodexTextGeneration.ts
const CODEX_MODEL = "gpt-5.3-codex";
// ...
yield* childProcess.spawn("codex", [
"prompt",
"--model",
CODEX_MODEL, // Hardcoded, ignores the model selected in the chat UI
// ...
Currently, the only workaround is to manually type the commit messages instead of relying on the auto-generate button, or use the chat to write the commit message and copy-paste it.
Another thought: considering cost and usage limits,
we should probably default to the lowest-cost model (based on the user's provider) that has an adequate context window.
Since this is just for generating commit messages, we don't necessarily need top-tier models for this specific task.
re: For example, if the user is using a heavy codex model (e.g. gpt-5.4 on xhigh) for chat,
we should use a lower-cost/faster model (e.g. gpt-5.4-mini, ⚠️ if it exists in the codex CLI!) for git commits.
changed the title [-][Bug]: Auto-generated commit messages use hardcoded `gpt-5.3-codex` model instead of respecting chat selection[/-][+][Bug]: Auto-generated commits fail when user lacks access to hardcoded gpt-5.3-codex model[/+]on Mar 17, 2026
Activity
AriajSarkar commentedon Mar 17, 2026
Another thought: considering
cost and usage limits,we should probably default to the
lowest-costmodel (based on theuser's provider) that has anadequate context window.Since this is just for generating
commit messages, we don't necessarily needtop-tier modelsfor this specific task.re: For example, if the user is using a heavy codex model (e.g. gpt-5.4 on xhigh) for chat,
we should use a lower-cost/faster model (e.g. gpt-5.4-mini,
⚠️ if it exists in the codex CLI!) for git commits.fix(git): use user-selected model for auto-generated commit messages
[-][Bug]: Auto-generated commit messages use hardcoded `gpt-5.3-codex` model instead of respecting chat selection[/-][+][Bug]: Auto-generated commits fail when user lacks access to hardcoded gpt-5.3-codex model[/+]UtkarshUsername commentedon Mar 18, 2026
Fixed by #1171