AI BENCHY 分类失败
指令遵循
答案错误
指令遵循
答案错误
看看哪些 AI 模型在 指令遵循 上最容易遇到 答案错误，更快找出薄弱点。 排序方式: 失败次数 ↑.
| 排名 | 模型 | 公司 | 答案错误 次数 | 分类得分 | 测试正确 | 响应时间（平均） |
|---|---|---|---|---|---|---|
| #15 | GPT-5.2 Chat none | OpenAI | 1 | 6.0 | 1/2 | 5.46s |
| #19 | GPT-5.3 Chat none | OpenAI | 1 | 9.0 | 1/2 | 3.29s |
| #20 | Gemini 3 Flash Preview none | 1 | 5.5 | 1/2 | 1.58s | |
| #25 | Claude Sonnet 4.6 none | Anthropic | 1 | 5.5 | 1/2 | 1.96s |
| #37 | Qwen3.5-Flash none | Qwen | 1 | 5.0 | 1/2 | 8.81s |
| #38 | Gemini 2.5 Flash none | 1 | 9.0 | 1/2 | 672ms | |
| #42 | Qwen3.5-35B-A3B none | Qwen | 1 | 5.0 | 1/2 | 809ms |
| #44 | GPT-5.4 none | OpenAI | 1 | 5.5 | 1/2 | 1.07s |
| #45 | Trinity Large Preview none | Arcee AI | 1 | 3.5 | 0/2 | 1.09s |
| #46 | Kimi K2.5 none | Moonshot AI | 1 | 5.5 | 1/2 | 2.67s |
| #47 | GPT-4o-mini none | OpenAI | 1 | 4.5 | 0/2 | 1.27s |
| #49 | GLM 4.7 Flash none | Z.ai | 1 | 5.5 | 1/2 | 888ms |
| #50 | Qwen3 Coder Next medium | Qwen | 1 | 4.5 | 0/2 | 7.34s |
| #51 | Mercury 2 none | Inception | 1 | 5.5 | 1/2 | 551ms |
| #52 | GLM 4.7 Flash medium | Z.ai | 1 | 5.0 | 1/2 | 2.97s |
| #54 | MiMo-V2-Flash none | Xiaomi | 1 | 5.5 | 1/2 | 857ms |
| #40 | Qwen3.5-122B-A10B none | Qwen | 2 | 4.5 | 0/2 | 585ms |
| #41 | Qwen3.5-27B none | Qwen | 2 | 4.5 | 0/2 | 815ms |
| #48 | Qwen3 Coder Next none | Qwen | 2 | 4.5 | 0/2 | 7.71s |
| #53 | Grok 4.1 Fast none | X AI | 2 | 10.0 | 0/2 | 923ms |
| #55 | LFM2-24B-A2B none | Liquid | 2 | 4.5 | 0/2 | 1.09s |