chatgpt plan shows up in /model but the session doesn\'t run as the selected model #860

Summary

you connect your chatgpt plan in command code with /connect, picking OpenAI ChatGPT/Codex. it shows up fine, and gpt-5.6-luna is selectable in /model with max effort. the session just doesn't run as that model. something else answers.

Expected Behavior

you'd expect the session to run on the chatgpt plan and behave as GPT-5.6 Luna. at minimum, whatever model answers should match the one named in the header.

Actual Behavior

the header says models: gpt-5.6-luna with max effort · taste-1 · openai (byok), but ask it "what model are you?" and the answer has nothing to do with the selection. two tabs with the same header gave two different wrong answers.\n\n- on gpt-5.6-luna: "I'm Command Code, currently running MiMo V2.6 Pro (xiaomi/mimo-v2.6-pro)." it only corrected itself to GPT-5.6 Luna after i pushed back with a screenshot.\n- on another gpt-5.6-luna tab: "I'm Command Code, a coding agent running on an Anthropic Claude model."\n\nso the chatgpt plan appears in the UI, but the requests don't look like they reach it. screenshots attached.

Steps to reproduce the issue

1. /connect, then pick OpenAI ChatGPT/Codex and sign in with the chatgpt plan.\n2. /model, then pick gpt-5.6-luna with max effort. the header now shows openai (byok).\n3. start a new session and ask "what model are you?"\n4. the reply names an unrelated model instead of the one you picked.

Command Code Version

1.54.0

Operating System

macOS

Terminal/IDE

Ghostty

Shell

zsh

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

screenshots attached: two tabs, both on gpt-5.6-luna with max effort · openai (byok), each returning a different wrong model identity. can't tell whether requests hit the chatgpt plan at all, or fall back to another model silently. happy to run whatever logging you need.