When I use “claude-4.6-opus-max-thinking” (without enabling the max mode), the number of requests will be consumed even when an error occurs. Is this a feature?
Hey @grunch!
Thanks for posting this. Saw another report of this this morning, and I’m escalating to the dev team!
We’ve rolled out a fix to avoid request consumption for null-token requests! If you (or another user) come across this again, please raise a new thread.