Toolathlon Verified Pass@3 (Anthropic Internal Harness): leaderboard
Toolathlon-Verified, the authors' final June 2026 release of 108 tool-use tasks with human-reviewed prompts, ground truth and evaluators, run on Anthropic's internal harness with adaptive thinking at max effort. Pass@3 is the percentage of tasks solved in at least one of three trials (higher is better). From Table 8.15.5.A of the Claude Fable 5.1 & Claude Mythos 5.1 System Card (September 2026).
Metric: Pass@3 (%). Source: www-cdn.anthropic.com. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 (Max) | 88 | #39 (Claude Opus 4.8) |
| 2 | Claude Opus 5 (Max) | 87 | #16 (Claude Opus 5) |
| 3 | Claude Sonnet 5 (Max) | 84.3 | #65 (Claude Sonnet 5) |
| 4 | Claude Opus 5.5 (Max) | 82.4 | #3 (Claude Opus 5.5) |
No result here: #6 GPT-6 Astra, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=toolathlon-verified-pass-3-anthropic-internal-harness · How It Works · Data refreshed daily, snapshot 2026-11-05.