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。 排序方式: 失败次数 ↑.
| 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） |
|---|---|---|---|---|---|---|
| #90 | Gemini 3.1 Flash Lite none | 1 | 6.4 | 9/21 | 1.06s | |
| #92 | Laguna M.1 medium | Poolside | 1 | 6.4 | 9/19 | 14.7s |
| #101 | Mimo V2 Omni none | Xiaomi | 1 | 6.0 | 8/21 | 2.44s |
| #103 | DeepSeek V4 Pro high | DeepSeek | 1 | 6.0 | 8/21 | 65.2s |
| #104 | Nemotron 3 Ultra 550b A55b none | NVIDIA | 1 | 6.0 | 8/21 | 2.27s |
| #106 | Grok 4.20 Beta none | X AI | 1 | 5.8 | 6/18 | 1.19s |
| #122 | GLM 4.7 Flash none | Z.ai | 1 | 5.5 | 6/21 | 2.86s |
| #125 | GPT-5.4 none | OpenAI | 1 | 5.5 | 7/21 | 1.42s |
| #128 | Qwen3.6 Flash none | Qwen | 1 | 5.4 | 7/21 | 1.60s |
| #133 | DeepSeek V3.2 none | DeepSeek | 1 | 5.2 | 6/21 | 13.8s |
| #139 | DeepSeek V4 Flash none | DeepSeek | 1 | 5.0 | 5/21 | 26.8s |
| #140 | Qwen3 Coder Next none | Qwen | 1 | 4.9 | 5/21 | 8.62s |
| #142 | Mistral Small 4 none | Mistral | 1 | 4.9 | 5/21 | 630ms |
| #143 | MiMo-V2.5 none | Xiaomi | 1 | 4.9 | 5/21 | 2.20s |
| #146 | Laguna Xs.2 none | Poolside | 1 | 4.8 | 5/19 | 806ms |