harness-evals (Terminal-Bench 2.1 subset): leaderboard
Claude Opus 5.5 and GPT-6.1 Sol each run 52 public Terminal-Bench 2.1 dev tasks in their own harness and in three third-party agent harnesses (Hermes Agent, OpenClaw, Pi) at effort high; the score is each task's pass share over two tries, averaged over the tasks. Unlike Terminal-Bench 2.1's own leaderboard it holds the model fixed to compare harnesses, on a 52-task subset.
Metric: Task-mean pass rate, 52 public Terminal-Bench 2.1 dev tasks, 2 attempts each (%). Source: github.com. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (High) | 85.6 | #10 (GPT-6.1 Sol) |
| 2 | Claude Opus 5.5 (High) | 81.7 | #5 (Claude Opus 5.5) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #6 GPT-6 Astra.
Interactive version: theaggregate.ai/benchmark?slug=harness-evals-terminal-bench-2-1-subset · How It Works · Data refreshed daily, snapshot 2026-11-13.