AI BENCHY 失败分析
未遵循指令 失败
看看哪些 AI 模型最常遇到 未遵循指令，让你在选择前先发现稳定性风险。
| 排名 | 模型 | 公司 | 未遵循指令 次数 | 分数 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #26 | Qwen3.6 Plus medium | Qwen | 1 | 7.9 | 14/21 | 30.7s |
| #28 | Gemini 2.5 Flash medium | 1 | 7.8 | 14/21 | 15.5s | |
| #32 | Gemini 3.5 Flash minimal | 1 | 7.7 | 14/21 | 1.57s | |
| #33 | Hy3 preview medium | Tencent | 1 | 7.7 | 14/21 | 16.3s |
| #39 | Qwen3.6 Flash medium | Qwen | 1 | 7.5 | 12/21 | 19.2s |
| #40 | Gemini 3.1 Flash Lite Preview medium | 1 | 7.5 | 13/21 | 3.96s | |
| #44 | Gemini 3.1 Flash Lite medium | 1 | 7.5 | 13/21 | 3.23s | |
| #46 | Qwen3.6 35B A3B medium | Qwen | 1 | 7.4 | 13/21 | 18.1s |
| #49 | Qwen3.5-Flash medium | Qwen | 1 | 7.4 | 12/21 | 63.3s |
| #50 | Gemini 3.1 Flash Lite Preview low | 1 | 7.4 | 13/21 | 2.77s | |
| #51 | Mimo V2 PRO medium | Xiaomi | 1 | 7.4 | 12/21 | 22.2s |
| #56 | MiMo-V2.5 medium | Xiaomi | 1 | 7.3 | 12/21 | 27.1s |
| #59 | GLM 5V Turbo medium | Z.ai | 1 | 7.2 | 11/21 | 23.1s |
| #64 | MiMo-V2-Flash medium | Xiaomi | 1 | 7.2 | 12/21 | 20.1s |
| #68 | Claude Opus 4.8 none | Anthropic | 1 | 7.0 | 12/21 | 3.47s |