MathArena ArXivMath (latest month)
Research-level questions with checkable answers taken from math papers posted to arXiv in the latest completed month, so they postdate training data. Only the newest month with at least five models is used, so all models face the same questions.
| Category | Math |
|---|---|
| Metric | Accuracy (%) |
| Source | MathArena (evaluator) |
| License | None stated |
| Benchmark page | matharena.ai |
| Status | Scored |
| Weight | 1.1 within Math |
| Fit | Discrimination 2.90, noise 0.47 |
Results
| # | Model | Provider | Result | Date |
|---|---|---|---|---|
| 1 | GPT-6.1 Sol Max | OpenAI | 90.4% | 2026-11-11 |
| 2 | Gemini 4 Argon | 83.0% | 2026-11-11 | |
| 3 | Claude Opus 5.5 High | Anthropic | 79.8% | 2026-11-11 |
| 4 | Kimi K3.1 Pro | Moonshot AI | 45.7% | 2026-11-11 |
| 5 | DeepSeek V4.1 Flash 1017 Max | DeepSeek | 38.3% | 2026-11-11 |