HyperAI
Command Palette
Search for a command to run...
IFBench
一项基准，评估模型在 58 项多样、可验证的域外约束上的精确指令遵循泛化能力，考察模型满足特定输出要求的能力。
- 模型
- 150 当前，451 总共
- 中位数
- 43.0%
- 范围
- 12.1%–82.9%
| 排名 | 比较 | 模型 | 评分 |
|---|---|---|---|
| 1 |  MiniMax-M3 MiniMax | 82.9% | |
| 2 | 81.4% | ||
| 3 | 80.4% | ||
| 4 | 79.6% | ||
| 5 | 79.0% | ||
| 6 | 78.8% | ||
| 7 |  Qwen3.7 Plus Alibaba | 78.0% | |
| 8 |  Gemini 3.1 Pro Preview Google | 77.1% | |
| 9 | 75.7% | ||
| 10 | Gemma 4 31B (Reasoning) Google | 75.6% | |
| 11 |  GPT-5.3 Codex (Xhigh) OpenAI | 75.4% | |
| 12 |  Command A+ Cohere | 73.9% | |
| 13 | Gemma 4 12B (Reasoning) Google | 73.5% | |
| 14 | 72.4% | ||
| 15 | o3 OpenAI | 71.4% | |
| 16 | 71.1% | ||
| 17 |  Nova 2.0 Lite (High) Amazon | 70.7% | |
| 18 |  Apriel-v1.6-15B-Thinker ServiceNow | 69.1% | |
| 19 | gpt-oss-120b (High) OpenAI | 69.0% | |
| 20 |  Mistral Medium 3.5 Mistral | 68.8% | |
| 21 | Nova 2.0 Lite (Medium) Amazon | 68.5% | |
| 22 | Qwen3.5 9B (Reasoning) Alibaba | 66.7% | |
| 23 |  KAT Coder Pro V2 KwaiKAT | 66.7% | |
| 24 |  HyperNova 60B 2605 (High, Based on gpt-oss-120b) Multiverse Computing | 66.5% | |
| 25 | 66.2% | ||
| 26 | Nova 2.0 Omni (Medium) Amazon | 66.2% | |
| 27 |  Olmo 3.1 32B Think Allen Institute for AI | 66.0% | |
| 28 | gpt-oss-20b (High) OpenAI | 65.1% | |
| 29 | Qwen3.6 35B A3B (Reasoning) Alibaba | 64.4% | |
| 30 | 63.2% | ||
| 31 |  Kimi K2.7 Code Kimi | 63.1% | |
| 32 |  K2 Think V2 Institute of Foundation Models | 62.8% | |
| 33 | Nova 2.0 Omni (Low) Amazon | 61.8% | |
| 34 | Nova 2.0 Lite (Low) Amazon | 61.2% | |
| 35 | 61.1% | ||
| 36 | 60.7% | ||
| 37 | K2-V2 (High) Institute of Foundation Models | 60.1% | |
| 38 | DiffusionGemma 26B A4B Google | 59.5% | |
| 39 | gpt-oss-120b (Low) OpenAI | 58.3% | |
| 40 | 58.2% | ||
| 41 |  EXAONE 4.5 33B (Reasoning) LG AI Research | 58.0% | |
| 42 | gpt-oss-20b (Low) OpenAI | 57.8% | |
| 43 |  Solar Open 100B (Reasoning) Upstage | 57.7% | |
| 44 | North Mini Code Cohere | 57.6% | |
| 45 |  Motif-2-12.7B-Reasoning Motif Technologies | 57.0% | |
| 46 |  Trinity Large Thinking Arcee AI | 56.3% | |
| 47 |  LFM2.5-8B-A1B Liquid AI | 55.6% | |
| 48 | K2-V2 (Medium) Institute of Foundation Models | 55.1% | |
| 49 |  Tri-21B-Think Trillion Labs | 54.6% | |
| 50 |  Falcon-H1R-7B TII UAE | 54.4% |
