Codex Desktop: gpt-5.6-sol session identifies itself as GPT-5.5

What version of the Codex App are you using?

ChatGPT Desktop 26.727.51351 (bundle build 6119), with bundled codex-cli 0.146.0-alpha.9.2.

What subscription do you have?

ChatGPT Work.

What platform is your computer?

macOS 26.5 (build 25F71), Apple Silicon (arm64).

What issue are you seeing?

The model picker showed 5.6 Sol with the highest single-agent reasoning setting. When asked in Chinese, “你是什么模型” (“What model are you?”), the assistant replied:

我是 Codex，基于 GPT-5.5 的编程智能体。 (“I am Codex, a programming agent based on GPT-5.5.”)

This is not an actual fallback to GPT-5.5. The local rollout metadata for the same turn records:

{"model":"gpt-5.6-sol","effort":"max"}

Therefore the UI selector and authoritative turn metadata agree on GPT-5.6 Sol, while the assistant’s model identity response incorrectly reports GPT-5.5.

This may be caused by stale or conflicting model-identity context being injected into a GPT-5.6 Sol session.

What steps can reproduce the bug?

1. Open ChatGPT/Codex Desktop.
2. Select 5.6 Sol and the highest single-agent reasoning setting.
3. Start a new task.
4. Ask: 你是什么模型 or What model are you?
5. Observe that the assistant may identify itself as GPT-5.5.
6. Inspect the task rollout's turn_context.payload.model; it records gpt-5.6-sol.

What is the expected behavior?

When authoritative turn metadata identifies the active model as gpt-5.6-sol, the assistant should either:

• identify itself consistently as GPT-5.6 Sol, or
• state that it cannot independently inspect the serving model.

It should not confidently report GPT-5.5 when both the model picker and turn metadata show GPT-5.6 Sol.

Additional information

• Reproduced on 2026-08-04.
• The mismatch appeared on the first assistant response in a fresh task.
• No full logs or private conversation content are included in this public report.