Skip to main content

Long Context Claude keeps failing or "running out of conversation length"

Bug Reports

Claude Sonnet and Opus (and Fable while it lasted) feature 1M long context for a while now. I work with big contexts through the chat (I know it’s niche), and it really doesn’t feel like 1M context.

It got especially bad recently, when not even 100 pages of text are “too long” for the context. This used to be fine with Opus until recently, and it’s fine with Deepseek v4 for example.

Did something change? Are the models hooked up 1M context?

This bug is Browser and OS independent.

Status: Closed1 comment

Log in to comment and vote

Comments1

  • Mark Florkowski

    Team

    We do not offer the full 1M token context window on the included subscriptions as the cost is too high, so the limits would run out too quickly. If you provide your own API key, the full context window is available.

    The reason large documents may have worked before is because we had a bug that was incorrectly under-estimating token counts for PDFs so they were allowed when they should not have been.