AI BENCHY 失败分析
答案错误 失败
看看哪些 AI 模型最常遇到 答案错误，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 | 答案错误 次数 | 分数 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #31 | GLM 5V Turbo medium | Z.ai | 3 | 7.8 | 11/18 | 15.0s |
| #33 | GLM 5.1 medium | Z.ai | 3 | 7.8 | 12/18 | 24.1s |
| #35 | MiMo-V2-Omni medium | Xiaomi | 3 | 7.7 | 11/18 | 16.8s |
| #41 | MiMo-V2-Flash medium | Xiaomi | 3 | 7.5 | 11/18 | 23.4s |
| #42 | Claude Sonnet 4.6 none | Anthropic | 3 | 7.4 | 11/18 | 4.98s |
| #47 | Grok 4.20 medium | X AI | 3 | 7.0 | 9/18 | 10.3s |
| #51 | Nemotron 3 Super medium | NVIDIA | 3 | 6.7 | 9/18 | 19.1s |
| #52 | Grok 4.1 Fast medium | X AI | 3 | 6.7 | 9/18 | 23.9s |
| #56 | Grok 4.20 Multi Agent Beta medium | X AI | 3 | 6.4 | 7/18 | 9.80s |
| #15 | Gemini 2.5 Flash medium | 4 | 8.2 | 13/18 | 12.1s | |
| #17 | Gemini 3.1 Flash Lite Preview medium | 4 | 8.2 | 13/18 | 3.74s | |
| #22 | Gemini 3.1 Flash Lite Preview low | 4 | 8.1 | 13/18 | 3.22s | |
| #29 | Gemini 3.1 Flash Lite Preview none | 4 | 7.9 | 12/18 | 1.30s | |
| #38 | GPT-5.4 Nano medium | OpenAI | 4 | 7.6 | 11/18 | 11.2s |
| #44 | GPT-5.4 Mini medium | OpenAI | 4 | 7.3 | 9/18 | 15.2s |