Codex app on Intel macOS cannot switch Fast mode; /fast is sent as normal message

Summary

On the latest Codex desktop app build available to me, I cannot switch Fast mode from the app. The previous visible Fast mode control appears to be gone, and typing /fast on in the Codex app composer is sent as a normal user message instead of being handled as a command.

I suspect this may be specific to the Intel macOS Codex app build, but it may also be related to how the app resolves Fast tier support when a custom OpenAI-compatible provider is configured.

Environment

• Codex app: 26.527.60818
• App build: 3437
• Bundle id: com.openai.codex
• Codex CLI bundled with app: codex-cli 0.136.0-alpha.2
• macOS: 13.3 (22E252)
• Machine architecture: x86_64 / Intel Mac
• App executable architecture: Mach-O x86_64
• Auth/config shape:
• model = "gpt-5.5"
• model_reasoning_effort = "high"
• service_tier = "fast"
• [features].fast_mode = true
• model_provider = "OpenAI" using a custom OpenAI-compatible base URL
• provider has requires_openai_auth = true
• sensitive tokens and private URLs are omitted/redacted

Steps to reproduce

1. Open Codex desktop app on Intel macOS.
2. Use a thread with gpt-5.5.
3. Try to find the Fast mode toggle in the app UI/settings.
4. Type /fast on in the app composer and send it.
5. Also try persisting Fast mode in ~/.codex/config.toml:

```toml
model = "gpt-5.5"
model_reasoning_effort = "high"
service_tier = "fast"

[features]
fast_mode = true
```

Expected behavior

• The Codex app should expose a visible way to toggle or inspect Fast mode, or /fast on / /fast status should be handled by the app as a command.
• If Fast mode is unavailable for this environment, the app should show a clear reason, for example unsupported model, unsupported auth mode, unsupported provider, unsupported build/platform, or account/plan limitation.

Actual behavior

• I cannot find a Fast mode toggle in the latest Codex app UI/settings.
• Sending /fast on from the Codex app composer is treated as a normal chat message rather than toggling Fast mode.
• Adding service_tier = "fast" and [features].fast_mode = true to ~/.codex/config.toml does not appear to make the current app thread switch to Fast mode.

Notes

• The current docs say Fast mode is available in the Codex app when signed in with ChatGPT, and can be persisted with service_tier = "fast" plus [features].fast_mode = true.
• The docs also say CLI /fast commands are catalog-driven and hidden when the active model does not advertise a Fast tier.
• If this is expected because custom model providers / proxy base URLs cannot use ChatGPT Fast credits, the app should surface that explicitly instead of silently removing the control.
• Similar but not identical issues I found:
• #19982
• #20061
• #23532
• #23998
• #24585
• #25776