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。
| 排名 | 模型 | 公司 | 答案错误 次数 | 分数 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #89 | GPT-4o-mini none | OpenAI | 13 | 4.9 | 4/18 | 2.00s |
| #91 | Mercury 2 none | Inception | 13 | 4.8 | 4/18 | 613ms |
| #95 | Grok 4.1 Fast none | X AI | 13 | 4.5 | 3/18 | 1.76s |
| #96 | GPT-5.4 Nano none | OpenAI | 13 | 4.5 | 2/18 | 1.40s |
| #76 | Kimi K2.5 none | Moonshot AI | 12 | 5.5 | 6/18 | 13.4s |
| #87 | Qwen3 Coder Next none | Qwen | 12 | 5.1 | 4/18 | 10.2s |
| #94 | MiMo-V2-Flash none | Xiaomi | 12 | 4.5 | 3/18 | 2.79s |
| #70 | Qwen3.5-122B-A10B none | Qwen | 11 | 5.7 | 6/18 | 3.69s |
| #78 | Trinity Large Preview none | Arcee AI | 11 | 5.3 | 5/18 | 5.07s |
| #83 | Mistral Small 4 none | Mistral | 11 | 5.2 | 5/18 | 665ms |
| #61 | Seed-2.0-Lite none | Bytedance Seed | 10 | 6.2 | 8/18 | 2.53s |
| #62 | Gemini 2.5 Flash none | 10 | 6.2 | 7/18 | 903ms | |
| #66 | GPT-5.4 none | OpenAI | 10 | 5.9 | 7/18 | 1.51s |
| #67 | Qwen3.5-27B none | Qwen | 10 | 5.9 | 6/18 | 1.74s |
| #74 | GLM 4.7 Flash none | Z.ai | 10 | 5.6 | 5/18 | 3.35s |