ArXivMath April-May 2026 Tools Unspecified (Anthropic Internal Harness): leaderboard
Accuracy on the April and May 2026 ArXivMath releases (81 problems), reported in Anthropic's Sonnet 5 card. Internal extended-thinking evaluation. Tools Unspecified. Sonnet 5 averages four attempts; the Opus 4.8 attempt count and tool setting are unspecified. Separate from copied MathArena leaderboard runs.
Metric: Accuracy (%). Source: www-cdn.anthropic.com. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 (Thinking) | 71 | #38 (Claude Opus 4.8) |
Interactive version: theaggregate.ai/benchmark?slug=arxivmath-april-may-2026-tools-unspecified-anthropic-internal-harness · How It Works · Data refreshed daily, snapshot 2026-10-16.