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） |
|---|---|---|---|---|---|---|
| #40 | Gemini 3.1 Flash Lite Preview medium | 7 | 7.5 | 13/21 | 3.96s | |
| #41 | Nemotron 3 Ultra 550b A55b medium | NVIDIA | 7 | 7.5 | 13/21 | 15.1s |
| #42 | GPT-5.2 medium | OpenAI | 3 | 7.5 | 13/21 | 16.9s |
| #44 | Gemini 3.1 Flash Lite medium | 7 | 7.5 | 13/21 | 3.23s | |
| #46 | Qwen3.6 35B A3B medium | Qwen | 4 | 7.4 | 13/21 | 18.1s |
| #47 | Grok Build 0.1 medium | X AI | 5 | 7.4 | 13/21 | 49.9s |
| #48 | Gemini 3 Flash Preview none | 8 | 7.4 | 13/21 | 1.65s | |
| #50 | Gemini 3.1 Flash Lite Preview low | 7 | 7.4 | 13/21 | 2.77s | |
| #52 | Claude Sonnet 4.6 medium | Anthropic | 4 | 7.4 | 13/21 | 17.1s |
| #19 | Seed-2.0-Lite medium | Bytedance Seed | 5 | 8.2 | 14/21 | 47.1s |
| #21 | GPT-5.4 medium | OpenAI | 5 | 8.0 | 14/21 | 22.3s |
| #22 | Step 3.7 Flash medium | Stepfun | 5 | 8.0 | 14/21 | 20.4s |
| #23 | GLM 5 Turbo medium | Z.ai | 4 | 8.0 | 14/21 | 23.0s |
| #24 | GPT-5.2 Chat none | OpenAI | 6 | 7.9 | 14/21 | 7.13s |
| #25 | Qwen3.5 Plus 2026-02-15 medium | Qwen | 4 | 7.9 | 14/21 | 73.8s |