AI BENCHY Category Failures
Instructions following: Wrong answer
Instructions following
Wrong answer
See which AI models are most likely to hit Wrong answer on Instructions following, so you can spot weak points faster.
Failure Reasons
| Rank | Model | Company | Wrong answer Count | Category Score | Tests Correct | Response Time (avg) |
|---|---|---|---|---|---|---|
| #32 | Gemini 3.5 Flash minimal | 1 | 6.4 | 1/2 | 893ms | |
| #48 | Gemini 3 Flash Preview none | 1 | 6.4 | 1/2 | 1.58s | |
| #55 | GLM 5.1 medium | Z.ai | 1 | 6.4 | 1/2 | 7.47s |
| #77 | Claude Sonnet 4.6 none | Anthropic | 1 | 6.5 | 1/2 | 1.96s |
| #85 | Gemma 4 31B none | 1 | 6.5 | 1/2 | 2.84s | |
| #88 | Qwen3.7 Plus none | Qwen | 1 | 6.3 | 1/2 | 929ms |
| #91 | GPT-5.5 none | OpenAI | 1 | 6.2 | 1/2 | 1.15s |
| #101 | Mimo V2 Omni none | Xiaomi | 1 | 6.5 | 1/2 | 4.26s |
| #102 | Gemma 4 26B A4B none | 1 | 6.3 | 1/2 | 690ms | |
| #106 | Grok 4.20 Beta none | X AI | 1 | 6.3 | 1/2 | 649ms |
| #108 | Qwen3.5-Flash none | Qwen | 1 | 6.3 | 1/2 | 8.81s |
| #109 | GLM 5V Turbo none | Z.ai | 1 | 6.5 | 1/2 | 1.97s |
| #111 | Owl Alpha medium | Openrouter | 1 | 6.5 | 1/2 | 10.2s |
| #113 | DeepSeek V4 Pro none | DeepSeek | 1 | 6.3 | 1/2 | 8.23s |
| #114 | Qwen3.5 Plus 2026-04-20 none | Qwen | 1 | 6.2 | 1/2 | 1.17s |