I’m posting this as feedback because I think we need confirmation from others before calling this a bug.
I just watched a charge increase over time as Claude, which was already done with code changes, was showing “summarizing context” with a spinner. The context was only ~25% full.
I force-quit Cursor to stop the charges! It as it had grown sluggish and wasn’t replying. I actually had a few performance and scrolling issues while using the chat in an Editor Tab for the first time. Probably not related to this though…
I’ve never gotten close to 8.8 million tokens before and I have done FAR bigger runs than this single file edit. Even the numbers above seem much bigger than they should have been. I’m sure Cursor staff can see my usage patterns.
Stats on those 2:
1 Like
Bumping this thread as I can not find it except in my own posts history. Was it removed?
high cost and also stuck in Summarizing chat context for 5mins each time
1 Like
I just noticed something odd with the token count during summarization. It jumped from 400k tokens to 1 million tokens, though I can’t confirm if it reached 3 or 5 million tokens. It seems to be caused by context summarization, but this is the first time I’ve seen context expand to nearly 3–5 million tokens, which I think is an anomaly.
Is there a way to flag this as an emergency bug? I just had a very simple refactor request take 2M tokens, which is about 10x more usage that I have seen in the past for this sort of thing. There is something going on here that is costing us serious chunks of usage when it happens. $1.39 to reorder text on a page is serious. I literally asked it to reorganize the code without changing the code itself. It did not need any wide source code context or complex thinking or tool calls.
This also happens with me. Summarizing chat context takes a long time and sometimes will error out and display a “try again” as if I lost connection
