Benchmark
ArxivMath · With tools
Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.
- Models measured
- 5
- Scores recorded
- 5
- Sources
- 2
- IQ weight
- No weight published
Score distribution
- 92.6%
- Mean
- 92.1%
- Median
- 3%
- Std dev
- 88.6%
- 10th percentile
- 96.9%
- 90th percentile
- 88.6%
- Min
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5.5Opus 5.5 System Card capability section · August 2026 · code sandbox · no internet · max · 4 attempts/problem· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96.9% |
| 2 | Claude Sonnet 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.2% |
| 3 | Claude Fable 5.1Opus 5.5 System Card capability section · August 2026 · code sandbox · no internet· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 92.1% |
| 4 | Claude Opus 5Opus 5.5 System Card capability section · August 2026 · code sandbox · no internet· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 90.4% |
| 5 | GPT-6 AstraOpus 5.5 System Card capability section · MathArena vendor harness · max · 2 attempts/problem · LLM judge· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 88.6% |
Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats