Automated page. Written and translated automatically by AI and checked against the sources by software; no human editor. SI-IQ is an index built from public benchmark results, not an IQ test. How it works
MATH Level 5: AI model results
MATH Level 5 is a Math benchmark run by Epoch AI. 86 models in our data have a published result.
It is part of the SI-IQ index (loading α = 1.27, chance level 0%).
- Evaluator
- Epoch AI
- Benchmark home page
- arxiv.org/abs/2103.03874
Results
| 1 | GPT-5 OpenAI | 98.1% | 121118–124 | 12 |
| 2 | GPT-5 mini OpenAI | 97.8% | 115112–118 | 12 |
| 3 | o4-mini OpenAI | 97.8% | 112109–115 | 10 |
| 4 | o3 OpenAI | 97.8% | 116113–119 | 10 |
| 5 | Claude Sonnet 4.5 Anthropic | 97.7% | 110107–113 | 12 |
| 6 | Qwen3-Max Alibaba | 97.1% | 106103–110 | 6 too little data |
| 7 | DeepSeek-R1 (May 2025)open DeepSeek | 96.6% | 104100–109 | 5 too little data |
| 8 | o3-mini OpenAI | 96.5% | 105101–108 | 9 |
| 9 | Claude Haiku 4.5 Anthropic | 96.4% | 10499–108 | 8 |
| 10 | Gemini 2.5 Pro (May 2025) Google DeepMind | 95.9% | 10498–110 | 3 too little data |
| 11 | Gemini 2.5 Pro (Mar 2025) Google DeepMind | 95.6% | 108102–113 | 3 too little data |
| 12 | GPT-5 nano OpenAI | 95.2% | 105102–109 | 11 |
| 13 | o1 OpenAI | 94.7% | 10399–106 | 6 too little data |
| 14 | DeepSeek-R1open DeepSeek | 93.1% | 9893–102 | 4 too little data |
| 15 | Claude 3.7 Sonnet Anthropic | 91.2% | 9995–103 | 5 |
| 16 | Grok-3 mini xAI | 90.9% | 10095–105 | 4 too little data |
| 17 | DeepSeek-R1-Distill-Llama-70Bopen DeepSeek | 89.9% | 9288–97 | 3 too little data |
| 18 | o1-mini OpenAI | 89.2% | 9287–96 | 4 too little data |
| 19 | Grok 3 xAI | 88.7% | 9691–100 | 4 too little data |
| 20 | GPT-4.1 mini OpenAI | 87.3% | 9188–95 | 11 |
| 21 | DeepSeek-R1-Distill-Qwen-14Bopen DeepSeek | 87.1% | 8984–93 | 3 too little data |
| 22 | Claude Opus 4 Anthropic | 85.0% | 9893–102 | 5 |
| 23 | Claude Sonnet 4 Anthropic | 84.4% | 9793–101 | 6 |
| 24 | Gemini 2.0 Pro Google DeepMind | 83.5% | 8983–94 | 3 too little data |
| 25 | GPT-4.1 OpenAI | 83.0% | 9187–94 | 11 |
| 26 | Gemini 2.0 Flash (Feb 2025) Google DeepMind, Google | 82.2% | 8782–91 | 3 too little data |
| 27 | o1-preview OpenAI | 81.6% | 8480–89 | 3 too little data |
| 28 | Mistral Medium 3 Mistral AI | 81.6% | 8681–90 | 4 too little data |
| 29 | GPT-4.5 OpenAI | 78.6% | 8783–92 | 3 too little data |
| 30 | DeepSeek-V3 (Mar 2025)open DeepSeek | 75.5% | 8581–90 | 5 too little data |
| 31 | Gemma 3 27Bopen Google DeepMind | 74.0% | 7772–81 | 5 |
| 32 | Llama 4 Maverickopen Meta AI | 73.0% | 8177–85 | 7 |
| 33 | Gemini 1.5 Pro (Sept 2024) Google DeepMind | 70.4% | 8075–85 | 3 too little data |
| 34 | GPT-4.1 nano OpenAI | 70.0% | 7672–80 | 9 |
| 35 | Qwen3-235B-A22Bopen Alibaba | 68.9% | 8580–91 | 3 too little data |
| 36 | Qwen2.5-Max Alibaba | 67.2% | 7873–82 | 4 too little data |
| 37 | Qwen Plus (Jan 2025) Alibaba | 65.3% | 7671–81 | 3 too little data |
| 38 | Phi-4open Microsoft Research | 64.9% | 7671–81 | 4 too little data |
| 39 | DeepSeek-V3open DeepSeek | 64.9% | 7773–82 | 5 too little data |
| 40 | Grok-2 (Dec 2024) xAI | 63.5% | 7570–80 | 3 too little data |
| 41 | Qwen2.5-72Bopen Alibaba | 63.2% | 7268–77 | 5 too little data |
| 42 | Llama 4 Scoutopen Meta AI | 62.3% | 7368–78 | 4 too little data |
| 43 | Gemini 1.5 Flash (Sep 2024) Google DeepMind | 61.9% | 7470–79 | 3 too little data |
| 44 | Claude 3.5 Sonnet (October 2024) Anthropic | 56.9% | 7671–80 | 4 too little data |
| 45 | Qwen-Turbo Alibaba | 56.2% | 6964–75 | 3 too little data |
| 46 | Qwen2.5-32Bopen Alibaba | 56.1% | 7166–76 | 3 too little data |
| 47 | GPT-4o (Aug 2024) OpenAI | 53.3% | 7368–77 | 9 |
| 48 | GPT-4o mini OpenAI | 52.6% | 6864–73 | 10 |
| 49 | Claude 3.5 Sonnet Anthropic | 51.7% | 7166–76 | 3 too little data |
| 50 | GPT-4o (May 2024) OpenAI | 51.0% | 7066–75 | 5 too little data |
| 51 | Mistral Large 2 (Nov 2024)open Mistral AI | 50.3% | 7166–76 | 4 too little data |
| 52 | GPT-4o (Nov 2024) OpenAI | 49.8% | 6964–73 | 5 |
| 53 | Llama 3.1-405Bopen Meta AI | 49.8% | 7065–75 | 3 too little data |
| 54 | Mistral Small 3.1open Mistral AI | 46.8% | 6863–73 | 3 too little data |
| 55 | GPT-4 Turbo (Apr 2024) OpenAI | 46.7% | 6864–73 | 4 too little data |
| 56 | Claude 3.5 Haiku Anthropic | 46.4% | 6762–72 | 4 too little data |
| 57 | Mistral Large 2 (Jul 2024)open Mistral AI | 44.8% | 6863–73 | 3 too little data |
| 58 | Mistral Small 3open Mistral AI | 44.8% | 6762–72 | 3 too little data |
| 59 | Llama 3.3 70Bopen Meta AI | 41.6% | 6661–71 | 6 |
| 60 | Gemini 1.5 Pro (May 2024) Google DeepMind | 40.7% | 6661–71 | 3 too little data |
| 61 | Qwen2-72Bopen Alibaba | 39.1% | 6458–69 | 2 too little data |
| 62 | Claude 3 Opus Anthropic | 37.5% | 6560–70 | 5 too little data |
| 63 | Llama 3.1-70Bopen Meta AI | 36.7% | 6358–68 | 3 too little data |
| 64 | Gemma 2 27Bopen Google DeepMind | 27.9% | 5752–63 | 3 too little data |
| 65 | WizardLM-2 8x22Bopen Microsoft | 25.7% | 5953–64 | 2 too little data |
| 66 | Gemini 1.5 Flash (May 2024) Google DeepMind | 25.1% | 5853–63 | 3 too little data |
| 67 | Mistral Large Mistral AI | 24.5% | 5651–62 | 3 too little data |
| 68 | Mixtral 8x22Bopen Mistral AI | 24.2% | 5549–61 | 2 too little data |
| 69 | GPT-4 (Jun 2023) OpenAI | 23.0% | 5751–63 | 4 too little data |
| 70 | Llama 3.1-8Bopen Meta AI | 22.9% | 5549–60 | 4 too little data |
| 71 | Hermes 2 Theta Llama-3 70Bopen Nous Research, Arcee AI | 22.7% | 5550–61 | 3 too little data |
| 72 | Llama 3-70Bopen Meta AI | 22.6% | 5752–62 | 3 too little data |
| 73 | Gemma 2 9Bopen Google DeepMind | 21.0% | 5347–59 | 3 too little data |
| 74 | Claude 3 Sonnet Anthropic | 18.2% | 5449–59 | 3 too little data |
| 75 | phi-3-medium 14Bopen Microsoft | 17.6% | 5145–57 | 2 too little data |
| 76 | GPT-3.5 Turbo (Nov 2023) OpenAI | 15.9% | 4943–55 | 2 too little data |
| 77 | Ministral 8Bopen Mistral AI | 14.9% | 4741–53 | 3 too little data |
| 78 | Claude 3 Haiku Anthropic | 14.9% | 5045–56 | 3 too little data |
| 79 | Ministral 3B Mistral AI | 14.4% | 4943–55 | 2 too little data |
| 80 | Claude 2 Anthropic | 11.7% | 4842–53 | 3 too little data |
| 81 | DBRXopen Databricks | 11.7% | 4640–52 | 2 too little data |
| 82 | GPT-3.5 Turbo (Jan 2024) OpenAI | 11.6% | 4742–53 | 5 too little data |
| 83 | Gemini 1.0 Pro Google DeepMind | 11.2% | 4640–52 | 3 too little data |
| 84 | Mistral NeMoopen Mistral AI | 10.8% | 4741–53 | 3 too little data |
| 85 | Mixtral 8x7Bopen Mistral AI | 10.0% | 4337–49 | 2 too little data |
| 86 | Llama 3-8Bopen Meta AI | 6.1% | 4034–46 | 3 too little data |
Data sources: Aider polyglot leaderboard data (Apache-2.0); Arena leaderboard dataset on Hugging Face (CC BY 4.0); Berkeley Function Calling Leaderboard, Gorilla project (Apache-2.0); Epoch AI, ‘Capabilities & benchmarking’ (CC BY 4.0); LiteLLM, model_prices_and_context_window.json (MIT); Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter (licences and changes).