Benchmarks

Canopy on public coding benchmarks

Our own runs of the released Canopy CLI, with the method, the failures and the limits written down next to each score.

  • Terminal-Bench 2.1

    Complete

    79.3%

    Mean of low and high effort

    Model
    DeepSeek-V4.1-Flash
    Runs
    10–12 Nov 2026
    See the full results
  • DeepSWE v1.1

    In progress

    No score yet. Results appear here when the run is complete.

Method

How we run these

  • Canopy runs as shipped: the released CLI, default prompt and tools, no benchmark-specific prompt or task-specific text.
  • We follow the public Harness Leaderboard’s rules: same model, low and high reasoning effort, three attempts per task, its scoring code, 95% Wilson intervals.
  • These are our own runs. Canopy is not listed on that leaderboard, and only its owner’s runs are its scores.
  • Every failed trial is classified, and failures caused by our own harness are counted as failures and listed.

Your next customer is about to call.

Start free, or book a demo.

Benchmarks: How Canopy Scores on Public Coding Tests