Hey, just to clarify the policy, nothing has officially changed.
The documented behavior is: if on-demand is off, and both pools Cursor Models and Other Models hit 100%, there is no guaranteed free usage. To keep going, you need to either turn on on-demand, upgrade, or wait for the reset in the next billing cycle. Any reduced-speed continuation you may have seen before isn’t a guaranteed part of the plan, so you shouldn’t rely on it as a permanent option. This is expected behavior, not a removed feature.
One more thing: if you do get reduced-speed continuation but requests get rejected when you select Auto or Composer 2.5, that’s a known issue with the default options in the model picker that we’re tracking. Workaround: pick Auto → Optimize for Cost, not Balance or Intelligence, or pick Composer 2.5 with Fast turned off. Related thread: Cannot use Auto or Composer 2.5 after Included Usage reaches 100%
If you expect to hit limits regularly, the most reliable option is to turn on on-demand with a team spend limit. That way you control costs and avoid a hard block. Let me know if you want help setting up the limits.