Mathematics · Benchmark profile
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.
Data verified 27 Jul 2026 · Methodology 1.8.0
Benchmark score on FrontierMath v2 Tiers 1-3
GPT-5.6 Sol leads at 89%, followed by GPT-5.6 Terra (84.9%) and GPT-5.6 Luna (78.6%).
10 ranked models · higher is better · labels show rank and score
| Model | Rank | Provider | Score |
|---|---|---|---|
| GPT-5.6 Sol | #1 | OpenAI | 89% |
| GPT-5.6 Terra | #2 | OpenAI | 84.9% |
| GPT-5.6 Luna | #3 | OpenAI | 78.6% |
| GPT-5.5 | #4 | OpenAI | 51.7% |
| GPT-5.5 Pro | #5 | OpenAI | 51% |
| GPT-5.4 Pro | #6 | OpenAI | 50% |
| GPT-5.4 | #7 | OpenAI | 47.6% |
| Claude Opus 4.8 | #8 | Anthropic | 47.2% |
| Claude Opus 4.7 | #9 | Anthropic | 43.8% |
| Claude Opus 4.6 | #10 | Anthropic | 40.7% |
One best score per model · higher is better
| Rank | Model | Provider | License | Evidence use | Score |
|---|---|---|---|---|---|
| #1 | GPT-5.6 Sol gpt-5.6-sol | OpenAI | closed | Reference only | 89% |
| #2 | GPT-5.6 Terra gpt-5.6-terra | OpenAI | closed | Estimated reference | 84.9% |
| #3 | GPT-5.6 Luna gpt-5.6-luna | OpenAI | closed | Estimated reference | 78.6% |
| #4 | GPT-5.5 gpt-5.5 | OpenAI | closed | Estimated reference | 51.7% |
| #5 | GPT-5.5 Pro gpt.5.5.pro | OpenAI | closed | Estimated reference | 51% |
| #6 | GPT-5.4 Pro gpt-5-4-pro | OpenAI | closed | Estimated reference | 50% |
| #7 | GPT-5.4 gpt-5.4 | OpenAI | closed | Reference only | 47.6% |
| #8 | Claude Opus 4.8 claude-opus-4-8 | Anthropic | closed | Reference only | 47.2% |
| #9 | Claude Opus 4.7 claude-opus-4-7 | Anthropic | closed | Reference only | 43.8% |
| #10 | Claude Opus 4.6 claude-opus-4-6 | Anthropic | closed | Reference only | 40.7% |
| #11 | GPT-5.2 gpt-5.2 | OpenAI | closed | Estimated reference | 40.7% |
| #12 | Muse Spark muse-spark | Meta | closed | Reference only | 39% |
| #13 | Gemini 3.5 Flash gemini-3.5-flash | closed | Estimated reference | 39.0% | |
| #14 | Kimi K2.6 kimi-k2.6 | Moonshot AI | open | Estimated reference | 39.0% |
| #15 | Gemini 3 Pro gemini-3-pro | closed | Reference only | 37.6% | |
| #16 | Gemini 3.1 Pro gemini-3-1-pro | closed | Estimated reference | 36.9% | |
| #17 | Gemini 3 Flash gemini-3-flash | closed | Reference only | 35.6% | |
| #18 | GLM-5.1 glm-5-1 | Z.AI | open | Reference only | 33.4% |
| #19 | Claude Sonnet 4.6 claude-sonnet-4-6 | Anthropic | closed | Reference only | 32.4% |
| #20 | GPT-5.1 gpt.5.1 | OpenAI | closed | Estimated reference | 31.0% |
| #21 | GPT-5.4 mini gpt-5.4-mini | OpenAI | closed | Estimated reference | 28.3% |
| #22 | Kimi K2.5 kimi-k2.5 | Moonshot AI | open | Reference only | 27.9% |
| #23 | Qwen3.6 Plus qwen3-6-plus | Alibaba Cloud | closed | Reference only | 26.2% |
| #24 | GPT-5.4 nano gpt.5.4.nano | OpenAI | closed | Reference only | 25.9% |
| #25 | o4-mini (high) o4-mini-high | OpenAI | closed | Estimated reference | 24.8% |
| #26 | Qwen 3.6 Max (preview) qwen3-6-max-preview | Alibaba Cloud | closed | Reference only | 23.1% |
| #27 | DeepSeek V3.2 deepseek-v3.2 | DeepSeek | open | Reference only | 22.1% |
| #28 | Kimi K2 kimi-k2 | Moonshot AI | closed | Reference only | 21.4% |
| #29 | Qwen3.5 Plus qwen3-5-plus | Alibaba Cloud | closed | Estimated reference | 21.0% |
| #30 | Claude Opus 4.5 claude-opus-4-5 | Anthropic | closed | Reference only | 20.7% |
| #31 | Grok 4 grok-4 | xAI | closed | Reference only | 19.7% |
| #32 | o3 o3 | OpenAI | closed | Reference only | 18.7% |
| #33 | GLM-5 glm-5 | Z.AI | open | Reference only | 16.4% |
| #34 | Gemini 2.5 Pro gemini-2.5-pro | closed | Reference only | 14.1% | |
| #35 | Claude Sonnet 4.5 claude-sonnet-4-5 | Anthropic | closed | Estimated reference | 13.5% |
| #36 | o1 o1 | OpenAI | closed | Estimated reference | 9.3% |
| #37 | Qwen3 235B 2507 (Reasoning) qwen3-235b-2507-reasoning | Alibaba Cloud | open | Estimated reference | 8.5% |
| #38 | Qwen3.5 Flash qwen3-5-flash | Alibaba Cloud | closed | Reference only | 6.2% |
| #39 | Claude Haiku 4.5 claude-haiku-4-5-20251001 | Anthropic | closed | Estimated reference | 5.9% |
| #40 | GPT-4.1 gpt-4-1 | OpenAI | closed | Reference only | 5.5% |
| #41 | Gemini 2.5 Flash gemini-2.5-flash | closed | Reference only | 4.8% | |
| #42 | GPT-4.1 mini gpt-4-1-mini | OpenAI | closed | Estimated reference | 4.5% |
| #43 | GLM-4.6 glm-4.6 | Z.AI | open | Reference only | 3.8% |
| #44 | Grok 3 [Beta] grok-3-beta | xAI | closed | Reference only | 3.8% |
| #45 | GLM-4.7 glm-4.7 | Z.AI | open | Reference only | 2.4% |
| #46 | Claude 3.5 Sonnet claude-3-5-sonnet | Anthropic | closed | Estimated reference | 2.1% |
| #47 | DeepSeek V3 deepseek-v3 | DeepSeek | open | Reference only | 1.7% |
| #48 | GPT-4.1 nano gpt-4-1-nano | OpenAI | closed | Estimated reference | 1.0% |
| #49 | Llama 4 Maverick llama-4-maverick | Meta | open | Reference only | 0.7% |
| #50 | GPT-4o gpt-4o | OpenAI | closed | Reference only | 0.3% |
| #51 | Llama 4 Scout llama-4-scout | Meta | open | Reference only | 0% |
The top of this snapshot is led by GPT-5.6 Sol at 89%; third place is 10.4 points behind. The top-10 spread is 48.3 points.
About FrontierMath v2 Tiers 1-3
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. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
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.
GPT-5.6 Sol by OpenAI currently leads with 89%.
51 models in the LuminaBench cohort have a qualifying score on this benchmark.
No. This benchmark is display-only and does not enter the overall Lumina composite.
Related
