[BUG] Long Context Beta API Error mid-long-context-session

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/27943
2. https://github.com/anthropics/claude-code/issues/27945
3. https://github.com/anthropics/claude-code/issues/27944

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

+1

+1

yes I also started getting this aprox 30 min ago, been using long-context for the last week otherwise.

Just started for myself at 1pm EST. Both Sonnet and Opus 1m context errors; can switch models to regular and continue but if conversation is too long from the 1m context, it errors Error: Error during compaction: Error: Conversation too long. Press esc twice to go up a few messages and try again. which is useless.

Pretty sure I just saw $40 get flushed down the toilet when a long running session just went 400 because of this crap. This will be fun, yay. Maybe stop messing with flagship products during the weekday?

Experienced the same issue

it's funny how they vibe code their own product and seem to work on production...

Experienced the same issue

+1
In the middle of two sessions. Have been using Opus 4.6[1m] all week.
If you don't want my money just say so Anthropic, you don't have to be coy with me.

I ran into the same issue — since the context window has already exceeded 200k, switching back to Opus 4.6 with its 200k context limit means it can't be compacted or rolled back, so I'm essentially forced to abandon most of the work I've already done.

This happened and then they posted on X that they identified distillation attacks by DeepSeek, Moonshot AI, and MiniMax with 24k fraudulent accounts. I bet they slammed shut the 1m context tokens to everyone on Maxx plans that's too much of a timing coincidence.

This happened and then they posted on X that they identified distillation attacks by DeepSeek, Moonshot AI, and MiniMax with 24k fraudulent accounts. I bet they slammed shut the 1m context tokens to everyone on Maxx plans that's too much of a timing coincidence.

thanks for the heads up, here's a related blog post Anthropic put up: https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks

Still, seems pretty easy to not screw over all your users, and/or mention on the official issue tracker if/when we will be getting our [1m] context back. Over the last week I adapted my entire workflow to utilize [1m] and now what, should I roll it back?

Edit: Other than timing I wouldn't think it's related to this issue. not enough overlap. Still It would be great to hear an ETA on this issue and if/when we will get it fixed

+1

Hi folks, we're actively investigating these errors. Sorry for the disruption.

We've temporarily paused an internal migration we were doing to our feature flag system that we think is the root cause here—could someone who was previously affected confirm whether or not their 1M context access is restored?

Restarted session and tried to pick up in the same conversation which had the error - no change for me.

Haven't tried a new session yet.

Thanks for the update—we're continuing to investigate, and feature flags are pretty sticky in cache. More reports/retries are welcome.

Same here — it doesn't work with a new session either.

If possible, sharing the request_id from failed requests would be a big help :)

If possible, sharing the request_id from failed requests would be a big help :)

API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"The long context beta is not yet available for this subscription."},"request_id":"req_011CYRhaDx7aNaVVRYLsZoao"}

Original error occurrences - two concurrent sessions: "req_011CYRPvaHwhbzEZCwwxD16E" & "req_011CYRQMtZEYgU7sa7StNmW2"

Latest retries for those sessions: "req_011CYRgswkfkWVTNRi9zoFMC" & "req_011CYRgxpRXuGwdErD45vbTG"

Tried /login and re-authorised in a new session: "req_011CYRhNxhY2uso48neD11wW"

API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"The long context beta is not yet available for this
subscription."},"request_id":"req_011CYRgqcdBFGqiTqgMYqJKV"}

If possible, sharing the request_id from failed requests would be a big help :)

It's working again — thanks so much!

confirm it too.

Works again!

Thanks folks. We're quite certain we were able to root cause.

Note that conversations that got interrupted by this bug should be resumable without any troubles. The work is not lost. claude --resume 😄

@stevenpetryk :( please fix vs-code's /context command, it won't work! ... And enable the option to select on plan approval: clear all context or not ;( which is only available on terminal.

@stevenpetryk :( please fix vs-code's /context command, it won't work! ... And enable the option to select on plan approval: clear all context or not ;( which is only available on terminal.

Thanks for the feedback! I'll take a look. If you haven't already, open a feature request issue and we can discuss in there.

working again for me