Mathematics benchmark
FrontierMath v2 (Tiers 1-3) leaderboard
FrontierMath v2 Tiers 1-3. Every model the catalog carries a published FrontierMath v2 (Tiers 1-3) value for, ranked by that value.
Epoch AI's corrected v2 core FrontierMath suite of private advanced mathematics problems. Models can reason iteratively and use Python; scores are pass rates on the private set.
FrontierMath v2 (Tiers 1-3) ranking
51 models with a published FrontierMath v2 (Tiers 1-3) value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Python-enabled iterative mathematical problem solving |
|---|---|---|---|
| 1 |  GPT-5.6 Sol | OpenAI | 89.000 |
| 2 | GPT-5.6 Terra | OpenAI | 84.900 |
| 3 | GPT-5.6 Luna | OpenAI | 78.600 |
| 4 | GPT-5.5 | OpenAI | 51.700 |
| 5 | GPT-5.5 Pro | OpenAI | 51.000 |
| 6 | GPT-5.4 Pro | OpenAI | 50.000 |
| 7 | GPT-5.4 | OpenAI | 47.600 |
| 8 |  Claude Opus 4.8 | Anthropic | 47.241 |
| 9 | Claude Opus 4.7 | Anthropic | 43.793 |
| 10 | Claude Opus 4.6 | Anthropic | 40.700 |
| 10 | GPT-5.2 | OpenAI | 40.700 |
| 12 |  Muse Spark | Meta | 39.000 |
| 13 |  Gemini 3.5 Flash | 38.966 | |
| 13 |  Kimi K2.6 | Moonshot AI | 38.966 |
| 15 | Gemini 3 Pro | 37.600 | |
| 16 | Gemini 3.1 Pro | 36.900 | |
| 17 | Gemini 3 Flash | 35.640 | |
| 18 |  GLM-5.1 | Z.AI | 33.448 |
| 19 | Claude Sonnet 4.6 | Anthropic | 32.400 |
| 20 | GPT-5.1 | OpenAI | 31.034 |
| 21 | GPT-5.4 mini | OpenAI | 28.280 |
| 22 | Kimi K2.5 | Moonshot AI | 27.900 |
| 23 |  Qwen3.6 Plus | Alibaba | 26.207 |
| 24 | GPT-5.4 nano | OpenAI | 25.860 |
| 25 | o4-mini (high) | OpenAI | 24.828 |
| 26 | Qwen 3.6 Max (preview) | Alibaba | 23.103 |
| 27 |  DeepSeek V3.2 | DeepSeek | 22.100 |
| 28 | Kimi K2 | Moonshot AI | 21.404 |
| 29 | Qwen3.5 Plus | Alibaba | 21.034 |
| 30 | Claude Opus 4.5 | Anthropic | 20.690 |
| 31 |  Grok 4 | xAI | 19.655 |
| 32 | o3 | OpenAI | 18.685 |
| 33 | GLM-5 | Z.AI | 16.434 |
| 34 | Gemini 2.5 Pro | 14.138 | |
| 35 | Claude Sonnet 4.5 | Anthropic | 13.495 |
| 36 | o1 | OpenAI | 9.310 |
| 37 | Qwen3 235B 2507 (Reasoning) | Alibaba | 8.481 |
| 38 | Qwen3.5 Flash | Alibaba | 6.207 |
| 39 | Claude Haiku 4.5 | Anthropic | 5.903 |
| 40 | GPT-4.1 | OpenAI | 5.517 |
| 41 | Gemini 2.5 Flash | 4.844 | |
| 42 | GPT-4.1 mini | OpenAI | 4.483 |
| 43 | GLM-4.6 | Z.AI | 3.819 |
| 44 | Grok 3 [Beta] | xAI | 3.793 |
| 45 | GLM-4.7 | Z.AI | 2.439 |
| 46 | Claude 3.5 Sonnet | Anthropic | 2.069 |
| 47 | DeepSeek V3 | DeepSeek | 1.724 |
| 48 | GPT-4.1 nano | OpenAI | 1.034 |
| 49 | Llama 4 Maverick | Meta | 0.690 |
| 50 | GPT-4o | OpenAI | 0.345 |
| 51 | Llama 4 Scout | Meta | 0.000 |
Evidence key: Observed
Rows are ordered by the value FrontierMath v2 benchmark hub published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.