/clear in interactive CLI ignores model from config.toml, falls back to gpt-5.4

What version of Codex CLI is running?

codex-cli 0.125.0

What subscription do you have?

Pro

Which model were you using?

Configured model = "gpt-5.5" with model_reasoning_effort = "xhigh" in ~/.codex/config.toml.

What platform is your computer?

macOS (Darwin 25.3.0, Apple Silicon)

What terminal emulator and version are you using (if applicable)?

zsh

What issue are you seeing?

After running /clear in an interactive CLI session, the model resets to gpt-5.4 instead of respecting the model = "gpt-5.5" setting in ~/.codex/config.toml.

Evidence

config.toml (top-level, before any [section]):

model = "gpt-5.5"
model_reasoning_effort = "xhigh"

codex exec correctly reads the config — a new non-interactive session resolves to the right model:

$ codex exec "Say hello"
model: gpt-5.5
reasoning effort: xhigh

But /clear in interactive mode creates threads with the wrong model. Thread history from state_5.sqlite:

| created_at | model |
|---|---|
| 2026-04-24 17:52 | gpt-5.5 ← manually selected in session |
| 2026-04-24 18:01 | gpt-5.4-mini ← after /clear (fast_mode was on) |
| 2026-04-24 18:17 | gpt-5.4-mini ← after /clear |

After disabling fast_mode via codex features disable fast_mode, /clear now falls back to gpt-5.4 instead of gpt-5.4-mini — but still not the configured gpt-5.5.

This confirms /clear does not re-read the model key from config.toml when creating a new thread. It appears to use an internal default (gpt-5.4, or gpt-5.4-mini when fast_mode is enabled).

What steps can reproduce the bug?

1. Set model = "gpt-5.5" in ~/.codex/config.toml
2. Start codex interactively — model shows as gpt-5.5 ✓
3. Run /clear
4. Check the model indicator — it now shows gpt-5.4 instead of gpt-5.5

What is the expected behavior?

/clear should re-read config.toml (or retain the configured model) and create the new thread with gpt-5.5, consistent with what codex exec and a fresh codex launch resolve to.

Additional information

• codex exec (non-interactive) correctly resolves to gpt-5.5 from config — so the config file is valid and parseable.
• Quitting and restarting codex (instead of /clear) also correctly picks up gpt-5.5.
• Possibly related: #13066 (model mismatch between selected and effective model in multi-agent sessions).