ChatGPT Work run stuck in RUNNING for 10+ hours after “Resume stream unavailable”

ChatGPT Bugs
, ,
1

ChatGPT Work run stuck in RUNNING for 10+ hours after “Resume stream unavailable”

Environment

  • ChatGPT Work

  • ChatGPT Business

  • GPT-5.6 Sol

  • Web browser + Windows desktop app

  • Firefox on web

  • Time zone: Europe/Warsaw

Issue

I encountered a Work run that appears to have become permanently stuck in the RUNNING state.

The run worked normally for approximately 112 minutes and made real progress. It then stopped while executing Step 2 of 7.

At that point the Windows desktop client showed:

Resume stream unavailable

However, the web client continued to show the Work run as still running.

The elapsed-time counter kept increasing, the Stop control remained available, but there was no further visible progress and no additional Work credits were consumed.

The two clients therefore appeared to disagree about the actual lifecycle state of the same run.

Symptoms

After the failure:

  • Step 1 was shown as completed.

  • Step 2 remained “In progress”.

  • Later steps remained “Waiting”.

  • The Stop button was still present.

  • The elapsed timer continued to increase.

  • No new output appeared.

  • No further progress occurred.

  • No additional Work quota/credits appeared to be consumed.

  • The desktop client reported that the stream could not be resumed.

  • The web client continued to present the run as actively running.

This state persisted for more than 10 hours.

Quota-reset test

There was an additional useful observation.

During the incident my available Work quota reached 0%.

Later, after the quota reset, the interface showed 100% Work availability again.

Despite that reset, the stuck run did not resume.

It remained in exactly the same state:

  • Step 1 completed

  • Step 2 “In progress”

  • remaining steps “Waiting”

  • Stop button present

  • timer still increasing

  • no new execution or output

This suggests that the run was not merely paused waiting for Work quota to become available again.

Current state

The run still appears as RUNNING in the web interface even though there is no observable execution progress.

The desktop client cannot reconnect to its output stream and reports “Resume stream unavailable”.

From the user side there is therefore no clear way to determine whether the backend task is still alive, has terminated, or has entered a stale lifecycle state.

Expected behavior

If the execution stream is interrupted, I would expect one of the following:

  1. the client reconnects to the existing Work run;

  2. the run enters an explicit recoverable/paused state and can be resumed;

  3. the run fails cleanly and is marked as failed/stopped;

  4. the interface exposes a way to retry or recover the run.

A run should not remain indefinitely presented as RUNNING when there is no observable execution and the client can no longer resume its stream.

Actual behavior

The run remains indefinitely in a stale RUNNING state.

The UI continues to show an active execution, including an increasing timer and Stop control, while no further work is performed and another client reports that the execution stream cannot be resumed.

This makes it difficult to tell whether it is safe to abandon the run, whether backend resources are still allocated, and whether a new Work run could conflict with the old one.

I have also opened a private OpenAI Support case and provided screenshots there.