Mathematics benchmark
FrontierMath v2 (Tier 4) leaderboard
FrontierMath v2 Tier 4. Every model the catalog carries a published FrontierMath v2 (Tier 4) value for, ranked by that value.
Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning.
FrontierMath v2 (Tier 4) ranking
45 models with a published FrontierMath v2 (Tier 4) 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-6 Astra | OpenAI | 97.600 |
| 2 | GPT-5.6 Sol | OpenAI | 83.000 |
| 3 | GPT-5.6 Terra | OpenAI | 68.300 |
| 4 | GPT-5.6 Luna | OpenAI | 58.500 |
| 5 | GPT-5.5 Pro | OpenAI | 39.600 |
| 6 | GPT-5.4 Pro | OpenAI | 37.500 |
| 7 | GPT-5.5 | OpenAI | 35.400 |
| 8 |  Claude Opus 4.8 | Anthropic | 31.250 |
| 9 | GPT-5.4 | OpenAI | 27.100 |
| 10 | Claude Opus 4.7 | Anthropic | 22.917 |
| 11 | Claude Opus 4.6 | Anthropic | 22.900 |
| 12 | GPT-5.2 | OpenAI | 18.800 |
| 13 |  Gemini 3 Pro | 18.750 | |
| 14 | Gemini 3.1 Pro | 16.700 | |
| 15 |  Muse Spark | Meta | 14.600 |
| 16 | Gemini 3.5 Flash | 14.583 | |
| 17 |  Kimi K2.6 | Moonshot AI | 14.580 |
| 18 |  GLM-5.1 | Z.AI | 12.500 |
| 18 | GPT-5.1 | OpenAI | 12.500 |
| 20 |  Qwen3.6 Plus | Alibaba | 8.333 |
| 21 | Claude Sonnet 4.6 | Anthropic | 8.300 |
| 22 | GPT-5.4 nano | OpenAI | 6.250 |
| 22 | o4-mini (high) | OpenAI | 6.250 |
| 24 | Kimi K2.5 | Moonshot AI | 4.200 |
| 25 | Claude Opus 4.5 | Anthropic | 4.167 |
| 25 | Claude Sonnet 4.5 | Anthropic | 4.167 |
| 25 | Gemini 2.5 Flash | 4.167 | |
| 25 | Gemini 2.5 Pro | 4.167 | |
| 25 | Gemini 3 Flash | 4.167 | |
| 25 | Qwen 3.6 Max (preview) | Alibaba | 4.167 |
| 31 | GLM-4.6 | Z.AI | 2.128 |
| 32 |  DeepSeek V3.2 | DeepSeek | 2.100 |
| 32 | GLM-5 | Z.AI | 2.100 |
| 34 | Claude Haiku 4.5 | Anthropic | 2.083 |
| 34 |  Grok 4 | xAI | 2.083 |
| 34 | o3 | OpenAI | 2.083 |
| 34 | Qwen3.5 Plus | Alibaba | 2.083 |
| 38 | GPT-5.4 mini | OpenAI | 2.080 |
| 39 | Claude 3.5 Sonnet | Anthropic | 0.000 |
| 39 | GLM-4.7 | Z.AI | 0.000 |
| 39 | GPT-4.1 | OpenAI | 0.000 |
| 39 | Grok 3 [Beta] | xAI | 0.000 |
| 39 | Kimi K2 | Moonshot AI | 0.000 |
| 39 | Qwen3 235B 2507 (Reasoning) | Alibaba | 0.000 |
| 39 | Qwen3.5 Flash | Alibaba | 0.000 |
Evidence key: Observed
Rows are ordered by the value FrontierMath Tier 4 v2 leaderboard 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.