What version of Codex CLI is running?

codex-cli 0.113.0

What subscription do you have?

Tested with ChatGPT Plus and ChatGPT Team accounts.

Which model were you using?

gpt-5.4

What platform is your computer?

Linux 6.6.87.2-microsoft-standard-WSL2 x86_64 unknown

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

Windows Terminal (WSL), xterm-256color

What issue are you seeing?

service_tier = "fast" appears ineffective or at least unclear in ChatGPT-authenticated Codex CLI websocket mode.

What I observed:

• Codex CLI uses websocket transport to wss://chatgpt.com/backend-api/codex/responses.
• After inspecting the first outbound response.create websocket frame, I don't see a service_tier field in the JSON payload.
• If service_tier: "fast" is forced directly into the websocket JSON request, the backend responds with an error like:

{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"Unsupported service_tier: fast"}}

• On direct websocket probes using the same ChatGPT auth tokens and account IDs, terminal response.completed.response.service_tier comes back as default on the tested Plus and Team accounts.

So from the outside it looks like:

• service_tier = "fast" in CLI config doesn't have an obvious effect in ChatGPT-auth mode, or
• fast requires a specific entitlement/plan, but that isn't clear from the current behavior.

What steps can reproduce the bug?

1. Use codex-cli 0.113.0 with ChatGPT auth.
2. Configure:

model = "gpt-5.4"
service_tier = "fast"

3. Run:

RUST_LOG=debug codex exec "Reply with OK only."

4. Observe that Codex CLI uses websocket transport to /backend-api/codex/responses.
5. Inspect the first outbound websocket response.create payload:
   service_tier is not present.
6. If service_tier: "fast" is injected directly into the websocket JSON request, the backend returns Unsupported service_tier: fast.
7. On direct websocket probes with the same ChatGPT auth/account, the terminal response.completed.response.service_tier is default on the tested accounts.

What is the expected behavior?

One of these should be true:

• service_tier = "fast" should actually request/select fast tier when the account is entitled to it, or
• Codex CLI should reject or warn that fast is unsupported in this auth mode, or
• The docs/config behavior should explicitly clarify which plans/entitlements are expected to return fast.

Additional information

• The websocket transport itself appears to work fine.
• The issue is specifically about how fast is requested, interpreted, or documented in ChatGPT-authenticated CLI mode.
• If needed, I can provide redacted websocket header / first-event captures.