Colab Pro+ sessions disconnecting after 3–5 hours (regression)

[anymore4u]

Hello,

I’m a Colab Pro+ subscriber, and I’m reporting what appears to be a regression in session stability.

One of the core features of Colab Pro+ is the ability to run notebooks for up to 24 hours continuously, even when the browser is closed. However, for the past 3–4 days, my sessions have been disconnecting and terminating after approximately 3–5 hours of execution.

This is happening consistently and without user interaction. The notebook is actively running, yet the runtime disconnects and the session is fully terminated.

This behavior makes it impossible to rely on Colab Pro+ for long-running workloads, which is precisely the reason I’m paying for this tier. If sessions cannot remain active for extended periods, the value proposition of Pro+ is effectively broken.

I’d appreciate clarification on whether:

• This is a known issue or recent change in policy
• There is an ongoing incident affecting Pro+ runtimes
• Or if there are specific limits currently being enforced that differ from the advertised behavior

Please let me know if logs, notebook examples, or timestamps would help investigate this issue.

Thank you.

[timechild666]

Same here on Pro+: sessions are dying early. Sometimes even ~1 hour and the runtime is fully terminated even while code is running. Getting worse.

[colaboratory-team]

Thanks for reporting the issue. Please submit feedback (Help > Send feedback from https://colab.research.google.com/), mentioning this GitHub issue number #5793, and let us know by replying here again. We'll look at logs for your assigned runtimes and investigate then.

[FrankieShih]

Same issue. Colab Pro+ randomly terminates sessions with error code KeyboardInterruption

[anymore4u]

Thanks for reporting the issue. Please submit feedback (Help > Send feedback from https://colab.research.google.com/), mentioning this GitHub issue number #5793, and let us know by replying here again. We'll look at logs for your assigned runtimes and investigate then.

I just sent it, I'll run it again...

[colaboratory-team]

Thanks, I got 2 for now, and they are tracked internally as b/476231891 and b/476159488.

[anymore4u]

Thanks, I got 2 for now, and they are tracked internally as b/476231891 and b/476159488.

I ran a new test and the same problem occurred. My notebook started up shortly before 10:50 p.m. on January 15, 2026, and ran until 6:07 a.m. on January 16, 2026.

My problem with all this is that the environment is unstable and keeps disconnecting the Google Drive I use on my laptop, so it takes about 30 minutes to stabilize, and when it does stabilize... this happens. It's frustrating because I already use a lot of computing units just to turn it on, and now that it's turned on, it's crashing on its own...

[Turkei]

Same thing is happening to me. I've sent feedback in Colab too.

[anymore4u]

@colaboratory-team Any news about the bug? I'm still suffering from it... Today my session lasted only 1 hour...

[nil1eeee]

Same here, I am also a pro+ sub. Notebkooks crash without any warning or error message - very similar to when a notebook exceeds the available ram, though I can guarantee this is not the issue here. This appears to be happening regularly since someday last week and as far as I can see this happens across all sorts of notebooks (which have been running fine and without problems in the past and for months). Haven't tested if this happens on runtimes with gpu support - for me it's just cpu runtimes (with or without additional ram)

[timechild666]

Still broken. Runtimes disconnect or terminate randomly.

[FrankieShih]

@colaboratory-team Any news about the bug? I'm still suffering from it... Today my session lasted only 1 hour...

It hasn't been fixed, I was interrupted this morning again.

[anymore4u]

@colaboratory-team Any news? We have several people with the same problem... We need a decision

[FrankieShih]

Thanks, I got 2 for now, and they are tracked internally as b/476231891 and b/476159488.

@colaboratory-team Any updates? I believe that there are multiple tickets (e.g., #5804) reporting the same issue.

This is a paid service, not an open-source project relying on voluntary maintenance. It is unacceptable that no root cause or resolution has been identified over the past two weeks. And I believe users affected by this bug should receive timely response and appropriate compensation for the wasted units.