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: 测试正确 ↑.
| 排名 | 模型 | 公司 | 答案错误 次数 | 分数 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #117 | Qwen3.5-35B-A3B none | Qwen | 12 | 5.6 | 7/21 | 3.37s |
| #118 | Qwen3.6 27B none | Qwen | 11 | 5.6 | 7/21 | 3.72s |
| #119 | Cobuddy medium | Baidu | 9 | 5.6 | 7/21 | 39.9s |
| #120 | Mimo V2 PRO none | Xiaomi | 11 | 5.6 | 7/21 | 2.27s |
| #121 | Owl Alpha none | Openrouter | 10 | 5.5 | 7/21 | 9.88s |
| #124 | Kimi K2.6 none | Moonshot AI | 11 | 5.5 | 7/21 | 13.3s |
| #125 | GPT-5.4 none | OpenAI | 13 | 5.5 | 7/21 | 1.42s |
| #127 | Grok 4.20 none | X AI | 10 | 5.4 | 6/18 | 1.11s |
| #128 | Qwen3.6 Flash none | Qwen | 12 | 5.4 | 7/21 | 1.60s |
| #100 | Grok Build 0.1 none | X AI | 7 | 6.0 | 7/19 | 28.7s |
| #101 | Mimo V2 Omni none | Xiaomi | 10 | 6.0 | 8/21 | 2.44s |
| #102 | Gemma 4 26B A4B none | 10 | 6.0 | 8/21 | 5.91s | |
| #103 | DeepSeek V4 Pro high | DeepSeek | 4 | 6.0 | 8/21 | 65.2s |
| #104 | Nemotron 3 Ultra 550b A55b none | NVIDIA | 12 | 6.0 | 8/21 | 2.27s |
| #105 | Nemotron 3 Super medium | NVIDIA | 5 | 5.8 | 8/21 | 32.0s |