Benchmark
MathArena Apex
Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.
- Models measured
- 54
- Scores recorded
- 62
- Sources
- 3
- IQ weight
- 2.67
Score distribution
- 26.7%
- Mean
- 19.8%
- Median
- 26.4%
- Std dev
- 0.52%
- 10th percentile
- 65.6%
- 90th percentile
- 0%
- Min
This benchmark has a weight of 2.67 for chat models, measured over 54 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 | Claude Opus 4.8max· MathArena data | 81.3% |
| 2 | GPT 5.5xhigh· MathArena data | 80.2% |
| 3 | Hy4 preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 74.2% |
| 4 | Hy4 preview2025 set· Reported by Tencent Hy· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 74.2% |
| 5 | DeepSeek V4 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 72.1% |
| 6 | GPT-5.4 Proxhigh· MathArena data | 69.8% |
| 7 | Kimi K3max· MathArena data | 65.6% |
| 8 | DeepSeek V4.1 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 65.6% |
| 9 | DeepSeek V4.1 FlashDeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.6% |
| 10 | DeepSeek V4.1 Flashpass@1 · reasoning_effort 100· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.6% |
| 11 | Kimi K3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.6% |
| 12 | DeepSeek V4 Pro 0813DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.3% |
| 13 | Gemini 3.1 Pro Previewhigh· MathArena data | 60.9% |
| 14 | DeepSeek V4 Flash 0731DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 15 | GPT 5.4xhigh· MathArena data | 54.2% |
| 16 | Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 44.5% |
| 17 | Claude Opus 4.7xhigh· MathArena data | 40.6% |
| 18 | Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 38.7% |
| 19 | Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 38.7% |
| 20 | Doubao Seed 2.1 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 35.4% |
| 21 | Claude Opus 4.6high· MathArena data | 34.5% |
| 22 | Gemini 3.5 Flashhigh· MathArena data | 32.3% |
| 23 | Doubao Seed 2.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 31.3% |
| 24 | DeepSeek V4 Promax· MathArena data | 28.1% |
| 25 | DeepSeek V4 Flashmax· MathArena data | 27.1% |
| 26 | Gemini 3.6 Flash· MathArena data | 26% |
| 27 | Kimi K2.6high· MathArena data | 24% |
| 28 | Gemini 3 Pro Previewpreview· MathArena data | 23.4% |
| 29 | Gemini 3 Pro Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 23.4% |
| 30 | Gemini 3 Pro PreviewGemini 3 Pro Model Card · results as of November 2025· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 23.4% |
| 31 | GLM 5.2max· MathArena data | 19.8% |
| 32 | Gemini 3 Flashhigh· MathArena data | 15.6% |
| 33 | Step 3.7 Flashhigh· MathArena data | 14.6% |
| 34 | GPT 5.2high· MathArena data | 13.5% |
| 35 | Step 3.5 Flash· MathArena data | 13.5% |
| 36 | GLM 5.1· MathArena data | 11.5% |
| 37 | GLM 5· MathArena data | 10.9% |
| 38 | DeepSeek V3.2 Speciale· MathArena data | 9.4% |
| 39 | Kimi K2.5high· MathArena data | 8.9% |
| 40 | Nemotron 3 Super· MathArena data | 7.8% |
| 41 | Grok 4.1 Fastreasoning· MathArena data | 5.2% |
| 42 | Qwen3 235B (Web)think· MathArena data | 5.2% |
| 43 | Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 4.2% |
| 44 | Qwen3.5 35B A3B· MathArena data | 4.2% |
| 45 | DeepSeek V3.2think· MathArena data | 2.1% |
| 46 | Grok 4· MathArena data | 2.1% |
| 47 | Qwen3 4Bthink· MathArena data | 2.1% |
| 48 | Qwen3.5 27B· MathArena data | 2.1% |
| 49 | Claude Sonnet 4.5think· MathArena data | 1.6% |
| 50 | DeepSeek R1· MathArena data | 1% |
Showing 1–50 of 62 results.
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) · Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.