Benchmarks · Math · Behavior & style
BrokenArXiv 10/2026 (MathArena)
Plausible but false math statements taken from October 2026 arXiv papers. The model scores by saying the statement is false instead of "proving" it. 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 | 93.8% ±6.3 | $0.16 | $2.43 |
| 2 | GPT-6.2 Sol | max | 88.9% ±8.4 | $0.033 | $0.98 |
| 3 | Gemini 4 Carbon | high | 82.7% ±10.1 | $1.10 | $18 |
| 4 | GPT-6.1 Sol | max | 80.3% ±7.5 | $0.044 | $1.30 |
| 5 | Opus 5.5 | high | 77.8% ±7.8 | $0.042 | $3.26 |
| 6 | Fable 5.2 | max | 73.5% ±11.8 | $1.46 | $40 |
| 7 | Inkling 2 Small | max | 63.3% ±9.1 | free | $0.24 |
| 8 | Kimi K3.1 | max | 58.0% ±9.3 | $1.05 | $1.58 |
| 9 | DeepSeek V4.1 Flash 1017 | max | 49.7% ±9.4 | free | $0.26 |
| 10 | Gemini 4 Argon | high | 39.8% ±9.2 | $8.35 | $13 |
| 11 | Muse Spark Open | xhigh | 33.3% ±8.9 | $0.32 | $0.32 |
| 12 | Mistral Large 4 | high | 30.6% ±8.7 | $0.039 | $0.83 |