Why isn’t GLM-5.2 configured with a 1M context window by default, and instead limited to 200K? kimi k3
How can I manually enable the 1M context window? Where is the toggle for this setting right now? Context window behavior is very opaque.
Custom models lack support for reasoning mode configuration and context window selection, making them very limited.
If custom model integration is provided, please avoid feature limitations. Otherwise, there is little incentive for users to adopt your software.
Hey, thanks for the request. This is working as intended, I’ll explain.
200K is the standard working context window in normal non-Max mode for models that have a larger physical window. GLM 5.2 and Kimi K3 follow this rule, so by default you see 200K instead of the full 1M. Other large-window models work the same way. The big window is an opt-in, not the default, since it costs more tokens, is slower, and retrieval quality gets worse on very large inputs.
To get the full 1M on GLM 5.2, turn on Max Mode in the model picker the Max toggle next to the model selection in chat. With Max Mode, Cursor uses the model’s full physical window. There used to be a separate issue on GLM 5.2 where early summarization kicked in around 200K in Max Mode, but that’s been fixed, and 1M now works correctly end-to-end. Same for Kimi K3.
About custom models setting reasoning mode and choosing the context window, I’ll pass that as feedback to the team. Let me know if 1M still doesn’t kick in after you enable Max Mode.
So this means the 1M context window is only available when Max Mode is enabled, and there is no independent option to turn the 1M window on or off for each model. I suggest adding a dedicated toggle for this. @deanrie