Benchmarks · Math
ArXivMath 10/2026 (MathArena)
Research-level math questions with a checkable final answer, drawn from arXiv papers of October 2026, too recent to be in training data. Share correct. From MathArena.
12 models · 12 results · data as of 2026-11-18 · matharena.ai ↗
Leaderboard
GPT-6.2 Astra max
Heavy line: best score at each price. Thin lines join one model's effort levels. Cost uses each model's most subsidized plan, or API price with no plan data. 4 more models under All models in settings.
All results
12 rows| # | Model | Effort | Score | Subscription $ / task | API $ / task |
|---|---|---|---|---|---|
| 1 | GPT-6.2 Astra | max | 95.7% ±5.0 | $0.15 | $2.31 |
| 2 | GPT-6.2 Sol | max | 95.7% ±5.0 | $0.020 | $0.60 |
| 3 | Fable 5.2 | max | 93.6% ±6.7 | $0.82 | $22 |
| 4 | Gemini 4 Carbon | high | 93.6% ±6.7 | $0.32 | $5.29 |
| 5 | GPT-6.1 Sol | max | 90.4% ±5.9 | $0.041 | $1.21 |
| 6 | Gemini 4 Argon | high | 83.0% ±7.6 | $2.17 | $3.26 |
| 7 | Opus 5.5 | high | 79.8% ±8.1 | $0.029 | $2.19 |
| 8 | Kimi K3.1 | max | 45.7% ±10.1 | $1.67 | $2.50 |
| 9 | DeepSeek V4.1 Flash 1017 | max | 38.3% ±9.8 | free | $0.30 |
| 10 | Inkling 2 Small | max | 27.7% ±9.0 | free | $0.50 |
| 11 | Muse Spark Open | xhigh | 19.1% ±8.0 | $1.00 | $1.00 |
| 12 | Mistral Large 4 | high | 18.1% ±7.8 | $0.061 | $1.28 |