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 |
|---|---|---|---|
| 51 | 53.5% | ||
| 52 |  Jamba Reasoning 3B AI21 Labs | 52.4% | |
| 53 | 52.0% | ||
| 54 |  Qwen3.5 4B (Reasoning) Alibaba | 52.0% | |
| 55 | 51.6% | ||
| 56 |  Doubao Seed Code ByteDance Seed | 51.4% | |
| 57 | Qwen3.5 Omni Plus Alibaba | 51.2% | |
| 58 | 50.8% | ||
| 59 |  Step3 VL 10B StepFun | 50.2% | |
| 60 |  Mi:dm K 2.5 Pro Korea Telecom | 49.3% | |
| 61 |  MiniCPM5-1B (Reasoning) OpenBMB | 49.3% | |
| 62 |  Mistral Small 4 (Reasoning) Mistral | 48.2% | |
| 63 | 47.1% | ||
| 64 |  Cogito v2.1 (Reasoning) Deep Cogito | 46.3% | |
| 65 |  LFM2 24B A2B Liquid AI | 45.9% | |
| 66 | 45.4% | ||
| 67 | 45.2% | ||
| 68 |  Ring-2.6-1T InclusionAI | 44.6% | |
| 69 | Magistral Small 1.2 Mistral | 44.4% | |
| 70 |  Gemma 4 E4B (Reasoning) Google | 44.2% | |
| 71 | LFM2.5-1.2B-Instruct Liquid AI | 43.8% | |
| 72 | 43.4% | ||
| 73 | Ring-flash-2.0 InclusionAI | 43.3% | |
| 74 |  LongCat Flash Lite LongCat | 43.1% | |
| 75 |  Llama 4 Maverick Meta | 43.0% | |
| 76 | Magistral Medium 1.2 Mistral | 43.0% | |
| 77 |  JT-35B-Flash China Mobile | 42.0% | |
| 78 | LFM2.5-1.2B-Thinking Liquid AI | 41.8% | |
| 79 |  Olmo 3 7B Think Allen Institute for AI | 41.5% | |
| 80 | 41.4% | ||
| 81 | 41.1% | ||
| 82 |  K2-V2 (Low) Institute of Foundation Models | 41.0% | |
| 83 | 40.5% | ||
| 84 | 40.5% | ||
| 85 | Qwen3 Next 80B A3B Instruct Alibaba | 39.7% | |
| 86 | Llama 4 Scout Meta | 39.5% | |
| 87 | Olmo 3.1 32B Instruct Allen Institute for AI | 39.2% | |
| 88 |  ERNIE 4.5 300B A47B Baidu | 39.1% | |
| 89 | 39.0% | ||
| 90 | 38.2% | ||
| 91 | Gemma 4 E2B (Reasoning) Google | 38.0% | |
| 92 | Qwen3.5 Omni Flash Alibaba | 38.0% | |
| 93 | 37.9% | ||
| 94 | Qwen3.5 9B (Non-reasoning) Alibaba | 37.8% | |
| 95 | 37.5% | ||
| 96 | JT-MINI China Mobile | 36.7% | |
| 97 |  Command A Cohere | 36.5% | |
| 98 |  EXAONE 4.0 32B (Reasoning) LG AI Research | 36.3% | |
| 99 |  Nova Premier Amazon | 36.2% | |
| 100 | 36.2% |
