Codex desktop / GPT-6.1: severe response latency across all sessions #53600

What version of the Codex App are you using?

26.1007.30912, build 20156. Verified from the installed desktop app's bundle metadata (com.openai.codex).

What subscription do you have?

Paid subscription; the specific plan was not collected.

What platform is your computer?

macOS 26.5.2, Apple Silicon / arm64.

Which model are you using?

GPT-6.1, as reported by the user. The reasoning setting for each affected session was not captured.

What issue are you seeing?

On October 13, 2026 (Asia/Taipei, UTC+8), the user reports that GPT-6.1 in the Codex desktop app is extremely slow across all sessions they use, rather than an isolated slow conversation.

The prolonged waiting disrupts interactive work and has made the user question the value of their paid subscription. The user explicitly requested that the assistant submit this report on their behalf.

This is a report of the user's experience. Exact time to first token, total response durations, onset of the slowdown, and a controlled comparison have not been collected. The report does not establish whether the bottleneck is model inference, request queueing, context processing, transport, or desktop orchestration.

What steps can reproduce the bug?

User-reported scenario; no deterministic minimal reproduction has been established:

1. Use GPT-6.1 in the Codex desktop app.
2. Submit ordinary interactive requests in different sessions.
3. Observe consistently severe perceived response latency across the sessions.

Please investigate the broad, cross-session latency issue and help identify what diagnostics are needed to isolate it.

What is the expected behavior?

Responses should arrive promptly enough for interactive work. If a request is queued, stalled, or blocked, the app should provide useful progress or an actionable recovery path.

Additional information

• Scope reported by the user: all their sessions are extremely slow.
• Report date: October 13, 2026, UTC+8.
• No /feedback upload/session ID is available for this submission.
• Potentially related reports: Codex desktop / GPT-6.1 Sol: apparent stall and delayed turn completion after tool work finishes #51656 and Windows desktop / GPT-6.1 Sol: slow responses (~22.5 effective tokens/s) and ~179 s compaction #52014. A shared root cause has not been established.
• No private project contents, credentials, conversation transcripts, or raw logs are attached.