Benchmark
MathArena 08/2026
Math benchmarks measure how well models solve maths problems. Scores are percentages.
All math benchmarksScores from MathArena data.
- Models measured
- 20
- Scores recorded
- 23
- Sources
- 1
- IQ weight
- 2.04
Score distribution
- 62.4%
- Mean
- 68.4%
- Median
- 23.5%
- Std dev
- 27.4%
- 10th percentile
- 88.1%
- 90th percentile
- 11.3%
- Min
This benchmark has a weight of 2.04 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | |
|---|---|---|
| 1 | GPT-6.2 Astramax· MathArena data | 93.5% |
| 2 | GPT-6.2 Solmax· MathArena data | 93.5% |
| 3 | Gemini 4 Carbonhigh· MathArena data | 88.1% |
| 4 | GPT-6.1 Solmax· MathArena data | 86.3% |
| 5 | GPT-6 Astramax· MathArena data | 81.9% |
| 6 | Claude Fable 5.1high· MathArena data | 79.8% |
| 7 | GPT-6 Solmax· MathArena data | 78.6% |
| 8 | Claude Fable 5.1low· MathArena data | 78% |
| 9 | Claude Opus 5.5high· MathArena data | 77.4% |
| 10 | GPT-6 Astralow· MathArena data | 74.7% |
| 11 | Qwen3.8 Maxmax· MathArena data | 69.6% |
| 12 | Muse Wondermax· MathArena data | 68.4% |
| 13 | Kimi K3.1 Promax· MathArena data | 64.3% |
| 14 | Kimi K3max· MathArena data | 61.9% |
| 15 | DeepSeek V4.1 Flash 1017max· MathArena data | 60.7% |
| 16 | DeepSeek V4.1 Flashmax· MathArena data | 54.8% |
| 17 | Gemini 4 Argonhigh· MathArena data | 54.2% |
| 18 | DeepSeek V4.1 Flash 1017max, DSH· MathArena data | 46.7% |
| 19 | Grok 4.7xhigh· MathArena data | 35.1% |
| 20 | Muse Spark Openxhigh· MathArena data | 29.2% |
| 21 | Gemini 3.8 Flashhigh· MathArena data | 27.4% |
| 22 | Muse Spark 1.3max· MathArena data | 20.2% |
| 23 | Mistral Large 4high· MathArena data | 11.3% |
Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.
Data credits: MathArena (ETH SRI / INSAIT), MIT — cite dekoninck2026matharena (licence)