AI BENCHY Category Failures
Puzzle Solving: Did not follow instructions
Puzzle Solving
Did not follow instructions
See which AI models are most likely to hit Did not follow instructions on Puzzle Solving, so you can spot weak points faster. Sort by: Response Time (avg) ↓.
Failure Reasons
| Rank | Model | Company | Did not follow instructions Count | Category Score | Tests Correct | Response Time (avg) |
|---|---|---|---|---|---|---|
| #38 | GPT-5.4 Nano medium | OpenAI | 2 | 4.0 | 0/3 | 3.65s |
| #72 | Hunter Alpha none | OpenRouter | 1 | 5.8 | 1/3 | 3.06s |
| #48 | Gemma 4 31B none | 1 | 5.5 | 1/3 | 2.95s | |
| #55 | MiMo-V2-Omni none | Xiaomi | 1 | 8.0 | 2/3 | 2.71s |
| #77 | GLM 5 Turbo none | Z.ai | 1 | 5.5 | 1/3 | 2.43s |
| #92 | Qwen3 Coder Next medium | Qwen | 2 | 3.1 | 0/3 | 2.30s |
| #58 | GLM 5V Turbo none | Z.ai | 1 | 5.3 | 1/3 | 2.22s |
| #73 | Mistral Small 4 medium | Mistral | 2 | 3.4 | 0/3 | 2.00s |
| #65 | MiMo-V2-Pro none | Xiaomi | 1 | 6.0 | 1/3 | 1.83s |
| #98 | LFM2-24B-A2B none | Liquid | 1 | 4.4 | 0/3 | 1.69s |
| #69 | Kimi K2.6 none | Moonshot AI | 2 | 3.4 | 0/3 | 1.66s |
| #66 | GPT-5.4 none | OpenAI | 1 | 5.6 | 1/3 | 1.52s |
| #75 | GLM 5.1 none | Z.ai | 1 | 5.7 | 1/3 | 1.48s |
| #67 | Qwen3.5-27B none | Qwen | 1 | 6.7 | 1/3 | 1.37s |
| #63 | Qwen3.5-35B-A3B none | Qwen | 1 | 3.9 | 0/3 | 1.34s |