자동화된 페이지입니다. AI가 자동으로 작성하고 번역했으며 소프트웨어가 출처와 대조해 검증했습니다. 사람 편집자는 없습니다. SI-IQ는 공개 벤치마크 결과로 만든 지수이며 IQ 검사가 아닙니다. 작동 방식
FrontierMath Tiers 1–3: AI 모델 결과
FrontierMath Tiers 1–3은(는) Epoch AI에서 실행하는 수학 벤치마크입니다. 당사 데이터에서는 모델 86개에 대해 발표된 결과가 있습니다.
SI-IQ 지수에 포함됩니다(적재량 α = 1.50, 우연 수준 0%).
- 평가 기관
- Epoch AI
- 벤치마크 홈페이지
- epoch.ai/frontiermath
결과
| 1 | GPT-6 Astra OpenAI | 93.7% | 150146–153 | 9 |
| 2 | GPT-6.1 Sol OpenAI | 93.7% | 148144–152 | 9 |
| 3 | Claude Opus 5.5 Anthropic | 91.2% | 144141–148 | 9 |
| 4 | Claude Fable 5.1 Anthropic | 90.2% | 140136–143 | 8 |
| 5 | GPT-6 Sol OpenAI | 89.8% | 140137–144 | 9 |
| 6 | GPT-5.6 Sol OpenAI | 89.1% | 140137–143 | 9 |
| 7 | Claude Sonnet 5.5 Anthropic | 88.8% | 140137–143 | 9 |
| 8 | GPT-5.5 Pro OpenAI | 87.7% | 139135–142 | 5 데이터 부족 |
| 9 | Claude Fable 5 Anthropic | 87.0% | 138134–141 | 9 |
| 10 | GPT-5.6 Terra OpenAI | 86.0% | 134131–137 | 9 |
| 11 | Gemini 4 Argon Google DeepMind | 86.0% | 140137–143 | 9 |
| 12 | Claude Opus 5 Anthropic | 85.6% | 136133–140 | 9 |
| 13 | GPT-5.5 OpenAI | 85.3% | 138135–141 | 10 |
| 14 | GPT-5.4 Pro OpenAI | 82.5% | 135131–138 | 5 데이터 부족 |
| 15 | GPT-5.6 Luna OpenAI | 82.1% | 130127–133 | 9 |
| 16 | Muse Wonder Meta AI | 82.1% | 136133–139 | 4 데이터 부족 |
| 17 | Claude Opus 4.8 Anthropic | 80.0% | 132129–135 | 9 |
| 18 | GPT-6 Luna OpenAI | 78.9% | 130127–133 | 9 |
| 19 | GPT-5.4 OpenAI | 78.6% | 131128–133 | 10 |
| 20 | Claude Haiku 5.5 Anthropic | 75.1% | 129126–132 | 9 |
| 21 | Qwen 3.8 Max Alibaba | 74.7% | 130127–133 | 7 데이터 부족 |
| 22 | Muse Spark 1.3 Meta AI | 74.4% | 128125–131 | 5 데이터 부족 |
| 23 | GPT-5.2 Pro OpenAI | 74.0% | 132128–135 | 3 데이터 부족 |
| 24 | Kimi K3오픈 Moonshot | 72.2% | 127124–130 | 9 |
| 25 | Gemini 3.7 Flash Google DeepMind | 71.6% | 131128–134 | 9 |
| 26 | Claude Opus 4.7 Anthropic | 70.2% | 127124–130 | 10 |
| 27 | Muse Spark Open오픈 Meta AI | 69.8% | 127124–130 | 7 |
| 28 | GLM-5.3오픈 Z.ai (Zhipu AI) | 68.8% | 125122–128 | 9 |
| 29 | Gemini 3.8 Flash Google DeepMind | 68.4% | 130127–133 | 9 |
| 30 | GPT-5.2 OpenAI | 67.4% | 125122–128 | 11 |
| 31 | DeepSeek V4.1 Flash오픈 DeepSeek | 67.4% | 128125–131 | 9 |
| 32 | Claude Opus 4.6 Anthropic | 66.0% | 123120–126 | 10 |
| 33 | Grok 4.6 xAI | 66.0% | 126123–129 | 9 |
| 34 | Claude Sonnet 5 Anthropic | 65.6% | 126123–129 | 9 |
| 35 | Qwen3.8 Max (0902) Alibaba | 65.6% | 127123–130 | 6 데이터 부족 |
| 36 | DeepSeek V4 Pro 0813오픈 DeepSeek | 64.6% | 128125–131 | 9 |
| 37 | Qwen3.7-Max Alibaba | 64.6% | 125122–128 | 10 |
| 38 | Gemini 3.5 Flash Google DeepMind | 62.8% | 126123–129 | 10 |
| 39 | Gemini 3.1 Pro Google DeepMind | 59.6% | 127124–130 | 10 |
| 40 | GLM-5.2오픈 Z.ai (Zhipu AI) | 59.2% | 121118–124 | 10 |
| 41 | Gemini 3.6 Flash Google DeepMind | 58.9% | 125122–128 | 9 |
| 42 | Mistral Large 4 Mistral AI | 58.6% | 122119–125 | 7 데이터 부족 |
| 43 | DeepSeek V4 Flash 0731오픈 DeepSeek | 57.5% | 124121–127 | 9 |
| 44 | Grok 4.5 xAI | 57.2% | 124121–127 | 6 데이터 부족 |
| 45 | Kimi K2.6오픈 Moonshot | 57.2% | 121118–124 | 10 |
| 46 | GLM-5.3-Flash오픈 Z.ai (Zhipu AI) | 55.8% | 120117–124 | 8 |
| 47 | GPT-5 Pro OpenAI | 55.8% | 124120–128 | 3 데이터 부족 |
| 48 | GPT-5 OpenAI | 55.4% | 121118–124 | 12 |
| 49 | Kimi K2.7 Code오픈 Moonshot | 54.0% | 118115–122 | 8 |
| 50 | Grok 4.7 xAI | 53.0% | 123120–127 | 7 데이터 부족 |
| 51 | GPT-5.4 Mini OpenAI | 51.2% | 117113–120 | 9 |
| 52 | Gemini 3 Flash Google DeepMind | 51.2% | 122119–125 | 10 |
| 53 | GPT-5 mini OpenAI | 46.7% | 115112–118 | 12 |
| 54 | Inkling-Small오픈 Thinking Machines | 46.3% | 118115–122 | 8 |
| 55 | DeepSeek-V4-Pro오픈 DeepSeek | 45.3% | 117113–120 | 10 |
| 56 | GPT-5.4 Nano OpenAI | 44.9% | 118115–121 | 9 |
| 57 | Grok 4.3 Beta xAI | 42.8% | 117114–121 | 6 데이터 부족 |
| 58 | Qwen 3.6 Plus Alibaba | 38.2% | 113109–117 | 7 데이터 부족 |
| 59 | GLM-5.1오픈 Z.ai (Zhipu AI) | 36.8% | 116112–120 | 8 |
| 60 | o4-mini OpenAI | 36.1% | 112109–115 | 10 |
| 61 | Qwen3.6 27B오픈 Alibaba | 35.1% | 115110–119 | 6 |
| 62 | Claude Opus 4.5 Anthropic | 34.4% | 115111–118 | 11 |
| 63 | Qwen3.7-Plus Alibaba | 34.4% | 115110–119 | 6 |
| 64 | Inkling오픈 Thinking Machines | 33.3% | 114110–117 | 8 |
| 65 | o3 OpenAI | 33.3% | 116113–119 | 10 |
| 66 | Qwen3.5 397B-A17B오픈 Alibaba | 31.2% | 114109–118 | 6 |
| 67 | Gemini 3.1 Flash-Lite Google | 27.7% | 109105–114 | 3 데이터 부족 |
| 68 | GPT-5.5 Instant OpenAI | 26.3% | 107103–111 | 4 데이터 부족 |
| 69 | Gemini 3.5 Flash-Lite Google DeepMind | 26.0% | 109106–113 | 8 |
| 70 | Gemini 2.5 Pro (Jun 2025) Google DeepMind | 24.6% | 108104–112 | 8 |
| 71 | Claude Sonnet 4.5 Anthropic | 23.9% | 110107–113 | 12 |
| 72 | Qwen 3.6 Flash Alibaba | 22.5% | 107103–111 | 6 데이터 부족 |
| 73 | Qwen 3.6 35B-A3B오픈 Alibaba | 20.4% | 111107–115 | 6 |
| 74 | GPT-5 nano OpenAI | 20.0% | 105102–109 | 11 |
| 75 | Qwen3.7 Flash Alibaba | 19.3% | 107103–112 | 4 데이터 부족 |
| 76 | Qwen3-Max Alibaba | 18.9% | 106103–110 | 6 데이터 부족 |
| 77 | o3-mini OpenAI | 18.6% | 105101–108 | 9 |
| 78 | Qwen 3.5 Flash (hosted 35B-A3B) Alibaba | 18.2% | 107103–111 | 6 데이터 부족 |
| 79 | o1 OpenAI | 14.7% | 10399–106 | 6 데이터 부족 |
| 80 | Claude Opus 4.1 Anthropic | 12.6% | 105102–109 | 6 데이터 부족 |
| 81 | GPT-4.1 mini OpenAI | 6.7% | 9188–95 | 11 |
| 82 | GPT-4.1 OpenAI | 6.0% | 9187–94 | 11 |
| 83 | GPT-4 Turbo (Apr 2024) OpenAI | 0.7% | 6864–73 | 4 데이터 부족 |
| 84 | GPT-4o mini OpenAI | 0.7% | 6864–73 | 10 |
| 85 | GPT-4o (Aug 2024) OpenAI | 0.4% | 7368–77 | 9 |
| 86 | GPT-3.5 Turbo (Jan 2024) OpenAI | 0.0% | 4742–53 | 5 데이터 부족 |
데이터 출처: 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 (라이선스 및 변경 사항).