Did not follow instructions Failure Ranking

AI BENCHY Failures

See which AI models run into Did not follow instructions most often, so you can spot reliability risks before choosing one. Sort by: Score ↑.

Models Shown

Total Failures

180

Most Affected Model

Categories

In category Puzzle Solving75 In category General Intelligence58 In category Instructions following22 In category Anti-AI Tricks17 In category Tool Calling6 In category Coding2

Rank	Model	Company	Did not follow instructions Count	Score	Tests Correct	Response Time (avg)
#11	Gemini 3.1 Flash Lite Preview high	Google	1	8.4	12/16	68.8s
#10	Qwen3.5-27B medium	Qwen	2	8.4	13/18	53.0s
#9	Qwen3.6 Plus Preview medium	Qwen	1	8.5	13/17	13.9s
#7	GPT-5.3-Codex medium	OpenAI	2	8.6	13/18	15.4s
#6	Seed-2.0-Lite medium	Bytedance Seed	2	8.6	13/18	30.4s

←

1 2 3 4 5 6

Did not follow instructions Failures