Cursor's Grok 4.5 High Drastically Increases Token Usage Per Turn

NickPassig · x · 2026-08-06

By examining Cursor's usage logs, the author found that the newly added Grok 4.5 High mode consumes significantly more tokens per turn compared to the default "Auto" mode.

This suggests that Auto might use some form of chunking or subagents in the background, whereas Grok 4.5 High processes all context at once. While the author praises the model's code quality (noting it's slightly slower), they express concerns about how these larger token turns will impact costs.

Original post →

More from coding & agent

coding & agent channel →