Cursor user reports Grok 4.5 unexpectedly burning API credits
A user ran a large build prompt in Cursor with the model set to Grok 4.5 at max settings, and noticed their balance steadily ticking down during the run. They had to add extra ad hoc credit just to let the job finish. Checking their usage log afterward, they found calls from several different models beyond Grok 4.5, including both Opus and Sonnet set to high.
Their expectation had been that Grok 4.5 would bill against first, but in practice the underlying agents invoked a mix of different models, which drove the instead. The user opened a support case with Cursor and is asking whether others have seen the same behavior.
Key points
- User ran a large build prompt in Cursor with Grok 4.5 set to max
- balance dropped continuously during the run, requiring extra ad hoc top-up to finish
- Usage log showed calls from multiple other models, including Opus and Sonnet both at high settings
- Expectation was that Grok 4.5 bills against first, but mixed drove instead
- User filed a support case with Cursor and is asking if others hit the same issue