AI BENCHY 分类失败
谜题求解
未遵循指令
谜题求解
未遵循指令
看看哪些 AI 模型在 谜题求解 上最容易遇到 未遵循指令，更快找出薄弱点。
| 排名 | 模型 | 公司 | 未遵循指令 次数 | 分类得分 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #36 | Mercury 2 medium | Inception | 2 | 1.7 | 0/3 | 934ms |
| #39 | gpt-oss-120b medium | OpenAI | 2 | 1.7 | 0/3 | 11.8s |
| #49 | GLM 4.7 Flash none | Z.ai | 2 | 3.7 | 0/3 | 1.00s |
| #50 | Qwen3 Coder Next medium | Qwen | 2 | 10.0 | 0/3 | 2.30s |
| #3 | GPT-5.3-Codex medium | OpenAI | 1 | 9.3 | 2/3 | 5.12s |
| #7 | Qwen3.5-27B medium | Qwen | 1 | 8.3 | 2/3 | 64.6s |
| #9 | GPT-5.4 medium | OpenAI | 1 | 7.0 | 2/3 | 9.13s |
| #13 | Step 3.5 Flash medium | Stepfun | 1 | 4.0 | 1/3 | 7.72s |
| #27 | GPT-5.2 medium | OpenAI | 1 | 7.0 | 2/3 | 5.47s |
| #28 | Kimi K2.5 medium | Moonshot AI | 1 | 4.0 | 1/3 | 45.4s |
| #30 | Grok 4.1 Fast medium | X AI | 1 | 4.0 | 1/3 | 8.08s |
| #32 | GPT-5 Mini medium | OpenAI | 1 | 4.3 | 1/3 | 14.1s |
| #34 | GPT-5 Nano medium | OpenAI | 1 | 4.0 | 1/3 | 19.8s |
| #37 | Qwen3.5-Flash none | Qwen | 1 | 1.3 | 0/3 | 5.90s |
| #38 | Gemini 2.5 Flash none | 1 | 4.7 | 1/3 | 576ms | |
| #41 | Qwen3.5-27B none | Qwen | 1 | 6.3 | 1/3 | 1.37s |
| #42 | Qwen3.5-35B-A3B none | Qwen | 1 | 1.7 | 0/3 | 1.34s |
| #44 | GPT-5.4 none | OpenAI | 1 | 4.0 | 1/3 | 1.52s |
| #52 | GLM 4.7 Flash medium | Z.ai | 1 | 10.0 | 0/3 | 12.9s |
| #55 | LFM2-24B-A2B none | Liquid | 1 | 3.3 | 0/3 | 1.69s |