GPT-5.6: 1M Context Mode and Fast Mode cannot be enabled simultaneously

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Hi Cursor team,
I’ve encountered an issue where GPT-5.6’s 1M context mode and Fast mode cannot be enabled at the same time. When I try to use both features together, they appear to be mutually exclusive — enabling one disables the other.
Steps to Reproduce:
Select GPT-5.6 as the model in Cursor
Enable 1M context mode (set model_context_window to 1000000 or enable via UI)
Attempt to also enable Fast mode
Observe that both modes cannot be active simultaneously
Expected Behavior:
Both 1M context mode and Fast mode should be able to work together, allowing users to benefit from the extended context window while still getting faster response times.
Actual Behavior:
The two modes appear to be mutually exclusive. Enabling one automatically disables the other.
Model: GPT-5.6
Additional Context:
This is limiting for workflows that require processing large codebases (benefiting from 1M context) while also needing fast iteration cycles (benefiting from Fast mode). Being able to combine both would significantly improve the experience for large-scale refactoring and long-running agent tasks.
Thanks for looking into this!

Steps to Reproduce

Open Cursor and start a new Chat session (Cmd+L / Ctrl+L)
Select GPT-5.6 as the model from the model dropdown in the chat input area
Enable MAX mode by clicking the MAX toggle in the chat input area (this unlocks the 1M context window)
Observe that the 1M context window is now active
Attempt to also enable Fast mode (e.g., via the model dropdown or the Fast variant toggle)
Observe that enabling Fast mode automatically disables MAX mode (1M context), or vice versa — the two toggles are mutually exclusive and cannot be active at the same time

Operating System

Windows 10/11

Version Information

cursor 3.17

For AI issues: which model did you use?

gpt5.6

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hey, thanks for the detailed report. This is actually intended behavior, not a bug.

Fast mode for GPT-5.6 is a tier with priority processing from OpenAI, and it’s only available on the standard 272k context. There’s no Fast option for the 1M long-context tier, so the two toggles switch between valid configurations instead of stacking. The same limitation applies to GPT-5.4 and GPT-5.5.

In practice:

  • For working with a large codebase, keep 1M context enabled.
  • For quick iterations, switch to Fast with the 272k window, it still fits a lot of code.

I’ve heard the idea of having both modes together, and I’ll pass it on as feedback. I can’t share an ETA for this right now since it also depends on what the model provider offers.