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） |
|---|---|---|---|---|---|---|
| #58 | GLM 5V Turbo none | Z.ai | 8 | 6.2 | 8/18 | 3.10s |
| #64 | DeepSeek V3.2 none | DeepSeek | 8 | 6.1 | 7/18 | 12.1s |
| #69 | Kimi K2.6 none | Moonshot AI | 8 | 5.8 | 7/18 | 2.05s |
| #73 | Mistral Small 4 medium | Mistral | 8 | 5.7 | 5/18 | 5.64s |
| #93 | GLM 4.7 Flash medium | Z.ai | 8 | 4.6 | 4/18 | 32.3s |
| #49 | Qwen3.5 Plus 2026-02-15 none | Qwen | 9 | 6.8 | 9/18 | 2.60s |
| #53 | GLM 5 none | Z.ai | 9 | 6.6 | 9/18 | 4.23s |
| #59 | Qwen3.5-Flash none | Qwen | 9 | 6.2 | 8/18 | 3.25s |
| #63 | Qwen3.5-35B-A3B none | Qwen | 9 | 6.1 | 7/18 | 3.82s |
| #65 | MiMo-V2-Pro none | Xiaomi | 9 | 6.0 | 7/18 | 2.39s |
| #72 | Hunter Alpha none | OpenRouter | 9 | 5.7 | 6/18 | 4.58s |
| #81 | Elephant medium | Openrouter | 9 | 5.2 | 5/18 | 1.27s |
| #82 | Grok 4.20 none | X AI | 9 | 5.2 | 5/18 | 1.11s |
| #85 | Elephant none | Openrouter | 9 | 5.2 | 5/18 | 1.23s |
| #92 | Qwen3 Coder Next medium | Qwen | 9 | 4.7 | 3/18 | 10.8s |