MathArena Apex
The hardest problems from recent competitions. MathArena.
- Published by
- MathArena
- Category
- Maths
- Index weight
- 0.5
- Models
- 46
- Data as of
- 9 Sept 2026
MIT — eth-sri/matharena.
- 1Claude Opus 4.881.3%
- 2GPT-5.580.2%
- 3GPT-5.4 Pro69.8%
- 4Kimi K365.6%
- 5Gemini 3.1 Pro60.9%
- 6GPT-5.454.2%
- 7Claude Opus 4.740.6%
- 8Claude Opus 4.634.5%
- 9Gemini 3.5 Flash32.3%
- 10DeepSeek V4 Pro28.1%
- 11DeepSeek V4 Flash27.1%
- 12Gemini 3.6 Flash26.0%
- 13Kimi K2.624.0%
- 14Gemini 3 Pro23.4%
- 15GLM-5.219.8%
46 of 46
| # | |||||
|---|---|---|---|---|---|
| 1 |  Claude Opus 4.8Anthropic | 81.3% | – | 59.4 | – |
| 2 |  GPT-5.5OpenAI | 80.2% | – | 63.7 | – |
| 3 | GPT-5.4 ProOpenAI | 69.8% | – | 61.1 | – |
| 4 |  Kimi K3Moonshot AIopen | 65.6% | – | – | – |
| 5 |  Gemini 3.1 ProGoogle | 60.9% | – | 63.9 | – |
| 6 | GPT-5.4OpenAI | 54.2% | – | 61.3 | – |
| 7 | Claude Opus 4.7Anthropic | 40.6% | – | 57.2 | – |
| 8 | Claude Opus 4.6Anthropic | 34.5% | – | 61.3 | – |
| 9 | Gemini 3.5 FlashGoogle | 32.3% | Gemini 3.5 Flash | 60.7 | – |
| 10 |  DeepSeek V4 ProDeepSeekopen | 28.1% | – | 56.1 | – |
| 11 | DeepSeek V4 FlashDeepSeekopen | 27.1% | – | 54.7 | – |
| 12 | Gemini 3.6 FlashGoogle | 26.0% | Gemini 3.6 Flash | 60.7 | – |
| 13 | Kimi K2.6Moonshot AIopen | 24.0% | – | – | – |
| 14 | Gemini 3 ProGoogle | 23.4% | – | 61.6 | – |
| 15 |  GLM-5.2Zhipu AIopen | 19.8% | GLM 5.2 | 57.4 | – |
| 16 | Gemini 3 FlashGoogle | 15.6% | Gemini 3 Flash | 56.6 | – |
| 17 |  Step 3.7 FlashStepFunopen | 14.6% | Step 3.7 Flash | 54.6 | – |
| 18 | GPT-5.2OpenAI | 13.5% | – | 55.5 | – |
| 19 | Step 3.5 FlashStepFunopen | 13.5% | Step 3.5 Flash | 54.9 | – |
| 20 | GLM-5.1Zhipu AIopen | 11.5% | GLM 5.1 | 60.1 | – |
| 21 | GLM-5Zhipu AIopen | 10.9% | GLM 5 | 59.0 | – |
| 22 | DeepSeek V3.2DeepSeekopen | 9.4% | – | 54.7 | – |
| 23 | Kimi K2.5Moonshot AIopen | 8.8% | – | 55.3 | – |
| 24 |  Nemotron 3 SuperNVIDIA | 7.8% | – | 44.9 | – |
| 25 |  Grok 4.1xAI | 5.2% | – | 55.9 | – |
| 26 |  Qwen3 235B A22BAlibabaopen | 5.2% | thinking, 2507 | 51.7 | – |
| 27 | Grok 4 Fast RxAI | 5.2% | Grok 4 Fast R | – | – |
| 28 | Qwen3.5 35B-A3BAlibabaopen | 4.2% | – | 53.4 | – |
| 29 | Grok 4xAI | 2.1% | Grok 4 | 57.2 | – |
| 30 | Qwen3.5 27BAlibabaopen | 2.1% | – | 55.4 | – |
| 31 | qwen3-4b-instruct-2507Alibabaopen | 2.1% | thinking, 2507 | 40.1 | – |
| 32 | GPT 5 AgentOpenAI | 2.1% | – | – | – |
| 33 | Claude Sonnet 4.5Anthropic | 1.6% | – | 52.8 | – |
| 34 | Qed NanoLM-Provers | 1.6% | – | – | – |
| 35 | GPT-5OpenAI | 1.0% | – | 54.5 | – |
| 36 | GPT-5.1OpenAI | 1.0% | – | 53.8 | – |
| 37 | GPT-5 miniOpenAI | 1.0% | – | 52.3 | – |
| 38 | gpt-oss-120bOpenAIopen | 1.0% | – | 50.5 | – |
| 39 | GLM-4.5Zhipu AIopen | 1.0% | GLM 4.5 | 50.2 | – |
| 40 | DeepSeek R1DeepSeekopen | 1.0% | 0528 | 48.9 | – |
| 41 | DeepSeek V3.2DeepSeekopen | 0.5% | – | 55.9 | – |
| 42 | Gemini 2.5 ProGoogle | 0.5% | Gemini 2.5 Pro | 54.1 | – |
| 43 | GLM-4.6Zhipu AIopen | 0.5% | GLM 4.6 | 47.6 | – |
| 44 | Qwen3.5-9BAlibabaopen | 0.5% | – | 47.5 | – |
| 45 | Qwen3 30B A3BAlibabaopen | 0.5% | – | 45.1 | – |
| 46 | Kimi K2 ThinkingMoonshot AIopen | 0.0% | – | 52.8 | – |