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） |
|---|---|---|---|---|---|---|
| #55 | GLM 5.1 medium | Z.ai | 4 | 7.3 | 12/21 | 33.7s |
| #56 | MiMo-V2.5 medium | Xiaomi | 5 | 7.3 | 12/21 | 27.1s |
| #57 | Step 3.7 Flash low | Stepfun | 8 | 7.3 | 12/21 | 15.7s |
| #58 | Gemini 3.1 Flash Lite Preview none | 7 | 7.2 | 12/21 | 1.21s | |
| #60 | Kimi K2.6 medium | Moonshot AI | 3 | 7.2 | 12/21 | 71.7s |
| #61 | Gemini 3.1 Flash Lite low | 9 | 7.2 | 12/21 | 1.89s | |
| #63 | GPT-5.3 Chat none | OpenAI | 7 | 7.2 | 12/21 | 6.34s |
| #64 | MiMo-V2-Flash medium | Xiaomi | 5 | 7.2 | 12/21 | 20.1s |
| #65 | Grok 4.20 medium | X AI | 6 | 7.1 | 12/21 | 27.7s |
| #68 | Claude Opus 4.8 none | Anthropic | 4 | 7.0 | 12/21 | 3.47s |
| #69 | Claude Opus 4.6 medium | Anthropic | 3 | 7.0 | 12/21 | 25.9s |
| #30 | Qwen3.5-27B medium | Qwen | 4 | 7.8 | 13/21 | 68.4s |
| #31 | DeepSeek V4 Flash high | DeepSeek | 6 | 7.7 | 13/21 | 45.8s |
| #36 | Qwen3.5 Plus 2026-04-20 medium | Qwen | 8 | 7.6 | 13/21 | 46.4s |
| #38 | Grok 4.3 medium | X AI | 5 | 7.6 | 13/21 | 47.5s |