HyperAI
Command Palette
Search for a command to run...
GPQA Diamond
GPQAの中でも最も難しい198問。PhDの専門家は65%の正答率を達成するが、熟練した非専門家はWebアクセスがあっても34%にとどまる。
- モデル
- 205 現在、 617 合計
- 中央値
- 71.3%
- 範囲
- 11.1%–96.3%
| ランク | 比較する | モデル | スコア |
|---|---|---|---|
| 101 |  Ring-flash-2.0 InclusionAI | 72.5% | |
| 102 | 71.4% | ||
| 103 |  K2 Think V2 Institute of Foundation Models | 71.3% | |
| 104 |  Qwen3.5 4B (Non-reasoning) Alibaba | 71.2% | |
| 105 |  MiniCPM5-2B OpenBMB | 70.2% | |
| 106 |  Mi:dm K 2.5 Pro Korea Telecom | 70.1% | |
| 107 |  Hermes 4 - Llama-3.1 70B (Reasoning) Nous Research | 69.9% | |
| 108 |  Nova 2.0 Omni (Low) Amazon | 69.9% | |
| 109 | Nova 2.0 Lite (Low) Amazon | 69.8% | |
| 110 |  Motif-2-12.7B-Reasoning Motif Technologies | 69.5% | |
| 111 | K2 Horizon 3.7B Institute of Foundation Models | 69.2% | |
| 112 |  Step3 VL 10B StepFun | 69.0% | |
| 113 |  gpt-oss-20b (High) OpenAI | 68.8% | |
| 114 | K2-V2 (High) Institute of Foundation Models | 68.1% | |
| 115 |  JT-MINI China Mobile | 67.6% | |
| 116 | gpt-oss-120b (Low) OpenAI | 67.2% | |
| 117 |  Llama 4 Maverick Meta | 67.1% | |
| 118 |  DiffusionGemma 26B A4B Google | 66.9% | |
| 119 |  Magistral Small 1.2 Mistral | 66.3% | |
| 120 |  Falcon-H1R-7B TII UAE | 66.1% | |
| 121 | 66.1% | ||
| 122 |  Solar Open 100B (Reasoning) Upstage | 65.7% | |
| 123 | 64.4% | ||
| 124 |  LongCat Flash Lite LongCat | 63.6% | |
| 125 | 63.6% | ||
| 126 |  Sarvam 30B (High) Sarvam | 63.3% | |
| 127 |  Celeris-1 Celeris | 63.1% | |
| 128 | 63.1% | ||
| 129 |  EXAONE 4.0 32B (Non-reasoning) LG AI Research | 62.8% | |
| 130 | Qwen3 Omni 30B A3B Instruct Alibaba | 62.0% | |
| 131 | 61.5% | ||
| 132 | gpt-oss-20b (Low) OpenAI | 61.1% | |
| 133 | 60.3% | ||
| 134 |  Tri-21B-Think Trillion Labs | 60.1% | |
| 135 | K2-V2 (Medium) Institute of Foundation Models | 59.8% | |
| 136 |  Olmo 3.1 32B Think Allen Institute for AI | 59.1% | |
| 137 | Llama 4 Scout Meta | 58.7% | |
| 138 | Gemma 4 E4B (Reasoning) Google | 57.6% | |
| 139 |  Phi-4 Microsoft | 57.5% | |
| 140 | Ministral 3 14B Mistral | 57.2% | |
| 141 | 57.2% | ||
| 142 | 57.1% | ||
| 143 | 57.0% | ||
| 144 | Nova Premier Amazon | 56.9% | |
| 145 | 55.9% | ||
| 146 |  LFM2.5-2.6B Liquid AI | 55.8% | |
| 147 | 55.7% | ||
| 148 | 55.5% | ||
| 149 | 54.9% | ||
| 150 | K2-V2 (Low) Institute of Foundation Models | 54.1% |
