ModelCards
← All benchmarks

Terminal-Bench Hard

Completing multi-step tasks in a real terminal: shell commands, builds, data wrangling and debugging, graded automatically.

Your workloadRAG over docs · 12K in · 600 out · 50K requests/mo · 30% cached
  1. 1Command A+ · Coherebest value
    25.0%

“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.

How each maker reported it

  • Command A+ launch post

    Cohere's own figures as stated in the post. MathVista and CharXiv rose from 73.5% and 46.9% on an earlier Command model that the text does not name; the chart caption suggests Command A Vision. Artificial Analysis Intelligence Index is a score, not a percentage.

Only each model's own maker's number is shown; scores a maker reports for rival models are left out.