Reaction
Gemini 4 reportedly now takes a “normal” number of steps on CursorBench
A user says its step counts are more uniform across reasoning efforts than those of other models.
TLDR
A user says Gemini 4 now takes a “normal” number of steps on CursorBench, calling it a fix. They also say its step counts are more uniform across reasoning efforts than those of other models.
Combined views
—
1 Source, first seen ago
