MathArena - ARXIVLEAN June: leaderboard
Lean-formalized research-level math problems from June arXiv papers, requiring machine-verifiable formal proofs rather than natural-language solutions.
Metric: Accuracy (%). Source: matharena.ai. Saturation forecast: Around October 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 65.22 | #6 (GPT-6 Astra) |
| 2 | Qwen 3.8 Max | 45.65 | #34 |
| 3 | GPT-5.6 Sol (Max) | 39.13 | #22 (GPT-5.6 Sol) |
| 4 | Claude Opus 5 (Max) | 32.61 | #16 (Claude Opus 5) |
| 5 | Muse Spark 1.3 | 28.26 | #31 |
| 6 | Claude Opus 4.8 (Max) | 23.91 | #39 (Claude Opus 4.8) |
| 7 | Gemini 3.7 Flash | 21.74 | #24 |
| 8 | Gemini 3.8 Flash | 19.57 | #30 |
| 9 | Claude Fable 5 (Max) | 15.22 | #20 (Claude Fable 5) |
| 10 | Gemini 3.1 Pro (Preview) | 10.87 | #64 |
| 11 | GLM-5.2 | 8.7 | #80 |
| 12 | Step 3.7 Flash | 0 | #247 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=matharena-arxivlean-june · How It Works · Data refreshed daily, snapshot 2026-11-05.