AI BENCHY Category Failures
Coding: Did not follow instructions
Coding
Did not follow instructions
See which AI models are most likely to hit Did not follow instructions on Coding, so you can spot weak points faster. Sort by: Failure Count ↑.
Failure Reasons
16/16
Filter models
No models match the current search and filters.
| Rank | Model | Company | Did not follow instructions Count | Category Score | Total Cost | Tests Correct | Response Time (avg) |
|---|---|---|---|---|---|---|---|
| #11 | Gemini 3.5 Flash medium | 1 | 7.9 | $0.582 | 2/3 | 12.6s | |
| #48 | Claude Opus 4.6 medium | Anthropic | 1 | 5.7 | $2.053 | 1/3 | 30.1s |
| #69 | Claude Opus 4.8 none | Anthropic | 1 | 5.5 | $0.539 | 1/3 | 3.29s |
| #70 | DeepSeek V4 Pro none | DeepSeek | 1 | 5.6 | $0.034 | 1/3 | 13.4s |
| #73 | GLM 5.2 none | Z.ai | 1 | 3.7 | $0.030 | 0/3 | 7.55s |
| #84 | Gemini 3.5 Flash minimal | 1 | 5.6 | $0.108 | 1/3 | 2.75s | |
| #127 | Kimi K2.6 none | Moonshot AI | 1 | 5.5 | $0.078 | 1/3 | 82.6s |
| #132 | Grok 4.1 Fast medium | X AI | 1 | 7.8 | $0.069 | 0/1 | 23.6s |
| #136 | MiMo-V2.5-Pro none | Xiaomi | 1 | 4.3 | $0.017 | 0/3 | 1.41s |
| #139 | Qwen3.5 Plus 2026-04-20 none | Qwen | 1 | 3.9 | $0.032 | 0/3 | 1.69s |
| #145 | DeepSeek V3.2 none | DeepSeek | 1 | 3.1 | $0.017 | 0/3 | 14.5s |
| #157 | Laguna M.1 medium | Poolside | 1 | 1.5 | $0.033 | 0/1 | 35.6s |
| #159 | Cobuddy medium | Baidu | 1 | 3.7 | $0.000 | 0/3 | 79.2s |
| #178 | MiMo-V2-Flash none | Xiaomi | 1 | 4.3 | $0.025 | 0/3 | 2.64s |
| #182 | Granite 4.1 8B none | IBM Granite | 1 | 4.5 | $0.003 | 0/3 | 775ms |