XNews

Gemini 3.8 Flash Uses Excessive Tokens and Steps in CursorBench 4.0

Gemini 3.8 Flash Uses Excessive Tokens and Steps in CursorBench 4.0

Analysis of CursorBench 4.0 data revealed that Gemini 3.8 Flash is an outlier in token and step usage. While its cost and score are mediocre (neither cheapest nor strongest), it uses vastly more tokens and steps than other models, analogous to a student who studies extremely hard but only achieves average results.

Source posts