I’m writing to report a highly frustrating issue I’ve been running into today. While working on my project, I keep hitting a sudden token limit within a single session/window. Here is the specific log output:
2026-02-19 16:47:25.665 [info] E0219 16:47:25.665532 85330 log.go:380] agent executor error: INVALID_ARGUMENT (code 400): invalid_request_error: prompt is too long: 201206 tokens > 200000 maximum
As an Ultra tier user, I have never encountered this hard restriction before. My workflow involves dealing with highly dense codebases—specifically, inspecting and optimizing complex, hand-written assembly kernels. This naturally requires a large context window. Right now, I can only load my files and fix one or two issues before the 200k cap is triggered.
Being forced to constantly open a new window just to continue working completely destroys my momentum and context. It is incredibly disappointing and makes the tool almost unusable for deep, sustained work.
Could the development team please look into this urgently? I’d like to know if this is a sudden bug, an unannounced restriction, or an issue with my Ultra subscription not registering the correct context length.