Study: top reasoning effort costs up to 42x more on tasks low effort already solves
zainhas · x · 2026-10-29
Researchers systematically tested models × reasoning effort levels × benchmarks to answer whether "xhigh" effort is worth the extra tokens. On tasks that low effort already solves, top effort costs up to 42× more — and you can't tell without rerunning. Takeaway: defaulting to max reasoning effort often burns tokens for nothing; pick effort levels per task.
More from coding & agent
- Rime launches Prism, a speech-to-speech model that kills dead air in voice agent calls — testingcatalog · 2026-10-29
- Solo Founders Podcast ep 34: WorkersIO founder on verifying mission-critical software — julianweisser · 2026-10-29
- Solo Founders Podcast ep 34: agents now write, review and test the code — julianweisser · 2026-10-29
- Scoping Is Now the Bottleneck in ML Projects, Says Louis Dorard — louisdorard · 2026-10-29
- Personal Agent Protocol draft 0.3 moves Mediated Sign In out of the core spec — yenkel · 2026-10-29
- From a bedroom Linux PC to cloud boxes: a year of iterating a remote Claude Code setup — bigaiguy · 2026-10-29