Reasoning Effort picker missing 'xhigh' for gpt-5.4 and other supported models

Resolved 💬 2 comments Opened Apr 2, 2026 by ben-vargas Closed Apr 2, 2026

Bug

The "Reasoning Effort" picker in the chat/prompt entry UI does not show the xhigh option for gpt-5.4 and other ChatGPT models that support it (e.g. Codex models), even though xhigh is defined in models.json and across all protocol layers (Rust, TypeScript, Python SDKs).

Only Low, Medium, and High appear in the picker.

Expected

Models whose supported_reasoning_levels include xhigh (like gpt-5.4) should show all four effort levels: Low, Medium, High, and XHigh.

Additional issue (macOS app)

On the macOS desktop app, selecting a different model when creating a new session does not appear to persist — the picker reverts to the default model after selection.

Environment

  • macOS desktop app and iOS app
  • gpt-5.4 model

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗