Best LLMs for Math
Ranked by Epoch Mock AIME accuracy (Epoch AI, epoch.ai, CC BY 4.0). 125 models have a published score. Where available, Epoch FrontierMath (T1–3) accuracy — a harder, research-level benchmark — is shown alongside for reference.
| Rank | Model | Developer | Released | Mock AIME accuracy | FrontierMath (T1–3) | Blended $/1M |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 | Anthropic | 2026-06-09 | 100.0% | 87.0% | 20.00 |
| 2 | Claude Fable 5.1 | Anthropic | 2026-09-01 | 100.0% | 90.2% | 20.00 |
| 3 | Claude Opus 5.5 | Anthropic | 2026-09-22 | 100.0% | 91.2% | 8.00 |
| 4 | Claude Sonnet 5.5 | Anthropic | 2026-09-28 | 100.0% | 88.8% | 4.00 |
| 5 | GPT-5.6 Sol | OpenAI | 2026-07-09 | 100.0% | 89.1% | 8.00 |
| 6 | GPT-6 Astra | OpenAI | 2026-09-04 | 100.0% | 93.7% | 20.00 |
| 7 | GPT-6 Sol | OpenAI | 2026-09-22 | 100.0% | 89.8% | 4.00 |
| 8 | GPT-6.1 Sol | OpenAI | 2026-09-29 | 100.0% | 93.7% | 4.00 |
| 9 | GPT-6.1 TerraNEW | OpenAI | 2026-10-26 | 100.0% | 90.9% | 2.00 |
| 10 | Gemini 4 ArgonNEW | 2026-10-13 | 100.0% | 86.0% | 4.00 | |
| 11 | Qwen3.8 Max 0902 | Alibaba | 2026-09-02 | 100.0% | 65.6% | 3.00 |
| 12 | GPT-5.6 Terra | OpenAI | 2026-07-09 | 99.7% | 86.0% | 4.50 |
| 13 | DeepSeek V4.1 Flash 1017NEW | DeepSeek | 2026-10-17 | 99.4% | 71.2% | 0.2625 |
| 14 | Muse WonderNEW | Meta | 2026-10-12 | 99.4% | 82.1% | 8.00 |
| 15 | Qwen3.8 Max | Alibaba | 2026-08-03 | 99.4% | 74.7% | 3.00 |
| 16 | Grok 4.6 | xAI | 2026-08-12 | 99.2% | 66.0% | 3.00 |
| 17 | Muse Spark 1.3 | Meta | 2026-09-02 | 99.2% | 74.4% | 2.00 |
| 18 | Claude Haiku 5.5NEW | Anthropic | 2026-10-07 | 98.9% | 75.1% | 0.20 |
| 19 | Claude Opus 5 | Anthropic | 2026-07-24 | 98.9% | 85.6% | 10.00 |
| 20 | GPT-6 Luna | OpenAI | 2026-09-22 | 98.9% | 78.9% | 0.20 |
| 21 | Gemini 3.8 Flash | 2026-09-02 | 98.9% | 68.4% | 1.50 | |
| 22 | DeepSeek V4 Pro 0813 | DeepSeek | 2026-08-13 | 98.6% | 64.6% | 0.99 |
| 23 | Claude Opus 4.8 | Anthropic | 2026-05-28 | 98.3% | 80.0% | 10.00 |
| 24 | DeepSeek V4.1 Flash | DeepSeek | 2026-09-10 | 98.3% | 67.4% | — |
| 25 | GPT-5.6 Luna | OpenAI | 2026-07-09 | 98.3% | 82.1% | 0.45 |
| 26 | Grok 4.7 | xAI | 2026-09-21 | 98.1% | 53.0% | 3.00 |
| 27 | Claude Opus 4.7 | Anthropic | 2026-04-16 | 97.8% | 70.2% | 10.00 |
| 28 | Grok 4.5 | xAI | 2026-07-08 | 97.8% | 57.2% | 3.00 |
| 29 | Gemini 3.7 Flash | 2026-08-13 | 97.2% | 71.6% | 1.50 | |
| 30 | Kimi K3 | Moonshot AI | 2026-07-16 | 97.2% | 72.2% | 6.00 |
| 31 | DeepSeek V4 Pro | DeepSeek | 2026-04-24 | 96.7% | 45.3% | 1.98 |
| 32 | Kimi K2.6 | Moonshot AI | 2026-04-20 | 96.1% | 57.2% | 1.7125 |
| 33 | Gemini 3.1 Pro Preview | 2026-02-19 | 95.6% | 59.6% | 4.50 | |
| 34 | Gemini 3 Flash Preview | 2025-12-17 | 95.6% | 51.2% | 1.125 | |
| 35 | Gemini 3.5 Flash | 2026-05-19 | 95.6% | 62.8% | 3.375 | |
| 36 | Kimi K2.7 Code | Moonshot AI | 2026-06-12 | 95.6% | 54.0% | 1.7125 |
| 37 | Qwen3.7 Max | Alibaba | 2026-05-21 | 95.6% | 64.6% | 3.75 |
| 38 | Claude Sonnet 5 | Anthropic | 2026-06-30 | 94.7% | 65.6% | 4.00 |
| 39 | DeepSeek V4 Flash 0731 | DeepSeek | 2026-07-31 | 94.4% | 57.5% | — |
| 40 | Gemini 3.6 Flash | 2026-07-21 | 94.2% | 58.9% | 1.50 | |
| 41 | GLM-5.3-Flash | Zhipu AI | 2026-08-26 | 93.9% | 55.8% | 0.1942 |
| 42 | GLM-5.1 | Zhipu AI | 2026-04-07 | 93.3% | 36.8% | 1.5686 |
| 43 | Grok 4.3 | xAI | 2026-04-17 | 93.3% | 42.8% | 1.5625 |
| 44 | Qwen3.6 Plus | Alibaba | 2026-04-02 | 93.3% | 38.2% | 1.125 |
| 45 | Qwen3.7 Plus | Alibaba | 2026-06-02 | 93.3% | 34.4% | 0.70 |
| 46 | Grok 4.20 (Reasoning) | xAI | 2026-03-09 | 92.2% | 44.9% | 1.5625 |
| 47 | Gemini 3 Pro Preview | 2025-11-18 | 91.4% | — | — | |
| 48 | Claude Opus 4.6 | Anthropic | 2026-02-05 | 91.1% | 66.0% | 10.00 |
| 49 | GLM-5.3 | Zhipu AI | 2026-08-14 | 91.1% | 68.8% | 1.9421 |
| 50 | Qwen3.6 27B | Alibaba | 2026-04-22 | 91.1% | 35.1% | 1.35 |
| 51 | Qwen3.6 Max Preview | Alibaba | 2026-04-20 | 91.1% | — | 2.925 |
| 52 | Inkling Small | Thinking Machines | 2026-07-30 | 90.0% | 46.3% | — |
| 53 | Muse Spark 1.1 | Meta | 2026-04-08 | 88.9% | — | 2.00 |
| 54 | GPT OSS 120B | OpenAI | 2025-08-05 | 88.9% | — | — |
| 55 | Inkling | Thinking Machines | 2026-07-15 | 88.9% | 33.3% | — |
| 56 | Qwen3.5 397B-A17B | Alibaba | 2026-02-15 | 88.9% | 31.2% | 1.35 |
| 57 | DeepSeek Reasoner | DeepSeek | 2025-12-01 | 87.8% | — | 0.315 |
| 58 | Qwen3 235B-A22B Thinking 2507 | Alibaba | 2025-07-25 | 86.7% | — | — |
| 59 | Qwen3.5 Plus | Alibaba | 2026-02-16 | 86.7% | — | 0.90 |
| 60 | Qwen3.6 35B-A3B | Alibaba | 2026-04-17 | 86.7% | 20.4% | 0.8438 |
| 61 | Qwen3.7 Flash | Alibaba | 2026-07-15 | 86.7% | 19.3% | 0.055 |
| 62 | GLM-5.2 | Zhipu AI | 2026-06-13 | 86.4% | 59.2% | 1.9421 |
| 63 | GPT-5.5 | OpenAI | 2026-04-23 | 84.4% | 85.3% | 11.25 |
| 64 | Qwen3.5 Flash | Alibaba | 2026-02-23 | 84.4% | 18.2% | 0.175 |
| 65 | Qwen3.6 Flash | Alibaba | 2026-04-27 | 84.4% | 22.5% | 0.5625 |
| 66 | Gemini 2.5 Pro | 2025-06-17 | 84.2% | 24.6% | 3.4375 | |
| 67 | GLM-4.7 | Zhipu AI | 2025-12-22 | 83.3% | — | 0.859 |
| 68 | Kimi K2 Thinking Turbo | Moonshot AI | 2025-11-06 | 83.1% | — | — |
| 69 | Claude Sonnet 4.6 | Anthropic | 2026-02-17 | 82.2% | — | 6.00 |
| 70 | Gemma 4 26B A4B IT | 2026-04-02 | 82.2% | — | — | |
| 71 | GLM-5 | Zhipu AI | 2026-02-12 | 80.0% | — | 1.1205 |
| 72 | Gemini 3.1 Flash Lite | 2026-05-07 | 80.0% | 27.7% | 0.5625 | |
| 73 | Gemma 4 31B IT | 2026-04-02 | 73.3% | — | — | |
| 74 | Gemini 3.5 Flash Lite | 2026-07-21 | 71.1% | 26.0% | 0.85 | |
| 75 | MiniMax-M3 | MiniMax | 2026-06-01 | 71.1% | — | 0.525 |
| 76 | Gemini 2.5 Flash Preview 05-20 | 2025-05-20 | 70.8% | — | — | |
| 77 | Qwen3.5 35B-A3B | Alibaba | 2026-02-23 | 70.0% | — | 0.6875 |
| 78 | Seed 1.6 | Bytedance Seed | 2025-10-15 | 69.4% | — | — |
| 79 | GPT-5.5 Instant | OpenAI | 2026-05-05 | 68.1% | 26.3% | 11.25 |
| 80 | Seed-OSS-36B-Instruct | Bytedance Seed | 2025-08-20 | 67.5% | — | — |
| 81 | Qwen3 32B | Alibaba | 2025-04-28 | 66.9% | — | 1.225 |
| 82 | DeepSeek R1 0528 | DeepSeek | 2025-05-28 | 66.4% | — | — |
| 83 | GPT OSS 20B | OpenAI | 2025-08-05 | 65.3% | — | — |
| 84 | Qwen3 30B A3B | Alibaba | 2025-04-28 | 62.8% | — | — |
| 85 | Qwen3 30B-A3B Instruct 2507 | Alibaba | 2025-07-28 | 62.2% | — | — |
| 86 | Qwen3.5 9B | Alibaba | 2026-02-23 | 61.7% | — | — |
| 87 | QwQ 32B | Alibaba | 2025-03-05 | 59.2% | — | — |
| 88 | GLM-4.7-Flash | Zhipu AI | 2026-01-19 | 58.3% | — | — |
| 89 | Qwen3 8B | Alibaba | 2025-04-29 | 56.1% | — | 0.31 |
| 90 | Qwen3.5 4B | Alibaba | 2026-03-02 | 55.8% | — | — |
| 91 | DeepSeek-R1-Distill-Qwen-32B | DeepSeek | 2025-01-20 | 55.6% | — | — |
| 92 | DeepSeek-R1 | DeepSeek | 2025-01-20 | 53.3% | — | — |
| 93 | DeepSeek Chat | DeepSeek | 2025-12-01 | 48.9% | — | 0.315 |
| 94 | Claude Opus 4.5 | Anthropic | 2025-11-24 | 48.1% | — | 10.00 |
| 95 | DeepSeek R1 0528 Qwen3 8B | DeepSeek | 2025-05-29 | 43.9% | — | — |
| 96 | Claude Opus 4 | Anthropic | 2025-05-22 | 42.2% | — | — |
| 97 | Claude Opus 4.1 | Anthropic | 2025-08-05 | 40.0% | — | — |
| 98 | DeepSeek V3 0324 | DeepSeek | 2025-03-24 | 37.8% | — | — |
| 99 | Claude Haiku 4.5 | Anthropic | 2025-10-15 | 35.8% | — | 2.00 |
| 100 | Claude Sonnet 4.5 | Anthropic | 2025-09-29 | 35.6% | — | 6.00 |
| 101 | Mistral Medium 3 | Mistral | 2025-05-07 | 32.2% | — | 0.80 |
| 102 | Magistral Small | Mistral | 2025-06-10 | 30.0% | — | — |
| 103 | Claude Sonnet 4 | Anthropic | 2025-05-22 | 28.9% | — | — |
| 104 | Magistral Small 1.2 | Mistral | 2025-09-18 | 28.1% | — | — |
| 105 | Gemma 3 27B IT | 2025-03-12 | 22.5% | — | — | |
| 106 | Claude Sonnet 3.7 | Anthropic | 2025-02-19 | 21.9% | — | — |
| 107 | Gemma 3 12B IT | 2025-03-12 | 16.7% | — | — | |
| 108 | DeepSeek-V3 | DeepSeek | 2024-12-26 | 15.8% | — | — |
| 109 | Claude Sonnet 3.5 v2 | Anthropic | 2024-10-22 | 8.5% | — | — |
| 110 | Mistral Large 2407 | Mistral | 2024-07-24 | 8.5% | — | — |
| 111 | Mistral Large 2.1 | Mistral | 2024-11-18 | 7.8% | — | — |
| 112 | Gemma 3 4B IT | 2025-03-12 | 7.5% | — | — | |
| 113 | GPT-4o (2024-08-06) | OpenAI | 2024-08-06 | 6.4% | 0.4% | 4.375 |
| 114 | GPT-4o (2024-05-13) | OpenAI | 2024-05-13 | 6.2% | — | 7.50 |
| 115 | GPT-4o (2024-11-20) | OpenAI | 2024-11-20 | 6.2% | — | 4.375 |
| 116 | Llama-3.3-70B-Instruct | Meta | 2024-12-06 | 5.1% | — | — |
| 117 | Claude Haiku 3.5 | Anthropic | 2024-10-22 | 4.3% | — | — |
| 118 | Llama-3.1-70B-Instruct | Meta | 2024-07-23 | 3.6% | — | — |
| 119 | Llama-3.2-90B-Vision-Instruct | Meta | 2024-09-25 | 2.6% | — | — |
| 120 | Qwen2.5 7B Instruct | Alibaba | 2024-09-19 | 2.5% | — | — |
| 121 | Claude Haiku 3 | Anthropic | 2024-03-13 | 1.8% | — | — |
| 122 | Llama-3.1-8B-Instruct | Meta | 2024-07-23 | 1.7% | — | — |
| 123 | Gemma 2 27B | 2024-07-13 | 1.4% | — | — | |
| 124 | Llama 3.2 1B Instruct | Meta | 2024-09-25 | 0.6% | — | — |
| 125 | Mistral 7B Instruct v0.3 | Mistral | 2024-05-22 | 0.3% | — | — |
FAQ
How is the Epoch Mock AIME accuracy measured?
It's the model's Epoch Mock AIME accuracy, one canonical row per model.
What is the top-ranked model for math right now?
Currently Claude Fable 5 by Anthropic, at 100.0% on this metric.
How is this different from the main leaderboard?
The main /leaderboard/ page ranks by Epoch's ECI. This page ranks only by epoch mock aime accuracy — useful when your workload is specifically about that skill.
Why do some models not appear here?
Only models with a published epoch mock aime accuracy are ranked — 716 of 841 tracked models have no published score here and are excluded rather than shown with a misleading zero.