Terminal-Bench 3.0

Label Shift MonitorML & Forecasting

Human baseline3h

TaxonomyML & ForecastingML & Forecasting

All trialsFable-5claude-codeTrial 04

100%

Trial reward

Fable-5Pass

Trial 04 — 1st of 8 by reward score. This model averages 100% across 8 trials on this task.

  • $0.91Cost
  • 2.5mTime
  • 10kOutput tokens

Passed — graded by execution verdict, the complete grading on this bench.