Terminal-Bench-Science 0.1 (Anthropic Internal Harness): leaderboard
Task success on 70 scientific research workflows in terminal environments, graded by hidden tests of produced artifacts. These Anthropic internal-harness results use max effort. The Opus 5.5 run averages ten trials per task with safeguards and its documented server-side fallback. Higher is better. Source: Claude Opus 5.5 system card.
Metric: Resolved (%). Source: www-cdn.anthropic.com. Saturation forecast: Around January 2028. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Max) | 58.7 | #3 (Claude Opus 5.5) |
| 2 | Claude Fable 5.1 (Max) | 52.6 | #9 (Claude Fable 5.1) |
| 3 | Claude Opus 5 (Max) | 29 | #12 (Claude Opus 5) |
No result here: #5 GPT-6 Astra, #7 Gemini 4 Argon, #8 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=terminal-bench-science-0-1-anthropic-internal-harness · How It Works · Data refreshed daily, snapshot 2026-10-16.