ArXivMath June 2026 No Tools (Mythos 5.1 System Card): leaderboard
ArXivMath June 2026: 49 research mathematics problems from arXiv abstracts, without tools. Claude models use max effort, averaged over four runs; GPT and Gemini comparators come from MathArena. Anthropic warns that Mythos 5.1 training data may overlap this problem set; contamination has not been ruled out.
Metric: Score (%). Source: www-cdn.anthropic.com. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Max) | 90.82 | #16 (Claude Opus 5) |
| 2 | GPT-5.6 Sol (Max) | 86.73 | #22 (GPT-5.6 Sol) |
| 3 | Claude Opus 4.8 (Max) | 82.65 | #39 (Claude Opus 4.8) |
| 4 | Gemini 3.1 Pro (Preview) | 65.99 | #64 |
| 5 | Grok 4.5 | 65.99 | #54 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=arxivmath-june-2026-no-tools-mythos-5-1-system-card · How It Works · Data refreshed daily, snapshot 2026-11-05.