Codex App regression after Apr 17 update: first message in a new thread sends immediately, follow-up messages stall on send button before inference starts

Resolved 💬 1 comment Opened Apr 18, 2026 by chinmayingalagavi Closed Apr 18, 2026

What version of Codex App are you using?

26.415.30602 (1773)

What subscription do you have?

Not specified.

What platform is your computer?

macOS

What issue are you seeing?

After updating the Codex App on April 17, 2026, the first message in a brand-new thread sends immediately, but every subsequent message in that same thread stalls for about 5 seconds before inference starts.

The delay happens on the send button itself: after clicking send, the button shows a loading spinner before the model begins thinking/streaming. This does not look like model latency. It appears to be in the app submit path before the request is actually handed off.

Important details:

  • This was not happening on April 16, 2026 before the update.
  • It reproduces in the Codex App only.
  • The same prompts do not lag in CLI.
  • The first message in a new thread is fast; follow-up messages lag.
  • The issue is not typing lag. Typing is fine. The lag starts only after clicking send.

What steps can reproduce the bug?

  1. Open Codex App 26.415.30602 (1773).
  2. Start a new thread.
  3. Send a first message, for example hi.
  4. Observe that it sends immediately.
  5. Send a second message in the same thread.
  6. Observe that the send button shows a loading spinner for about 5 seconds before any thinking/response activity begins.
  7. Repeat with more follow-up messages in the same thread; the lag persists.
  8. Run the same prompt flow in CLI; CLI does not reproduce the delay.

What is the expected behavior?

Follow-up messages in an existing thread should submit immediately, as they did before the update, without a multi-second stall on the send button before inference starts.

Additional information

This looks like a regression in the app's conversation continuation/submit path rather than model inference speed.

Possible areas to inspect:

  • follow-up thread state assembly
  • request serialization / context packaging after turn 1
  • thread metadata sync before submit
  • app-side UI blocking in the submit path

A visual indicator of the symptom is that the spinner appears on the send control itself before the run starts, rather than a normal model-thinking state after submit.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗