AI BENCHY Category Failures
Anti-AI Tricks: Wrong answer
Anti-AI Tricks
Wrong answer
See which AI models are most likely to hit Wrong answer on Anti-AI Tricks, so you can spot weak points faster. Sort by: Tests Correct ↓.
Failure Reasons
| Rank | Model | Company | Wrong answer Count | Category Score | Tests Correct | Response Time (avg) |
|---|---|---|---|---|---|---|
| #8 | Claude Opus 4.7 none | Anthropic | 1 | 8.3 | 3/4 | 2.12s |
| #11 | Claude Opus 4.7 medium | Anthropic | 1 | 8.3 | 3/4 | 1.85s |
| #13 | Grok 4.20 Beta medium | X AI | 1 | 8.7 | 3/4 | 3.16s |
| #15 | GPT-5.3-Codex medium | OpenAI | 1 | 8.7 | 3/4 | 4.16s |
| #19 | Seed-2.0-Lite medium | Bytedance Seed | 1 | 8.3 | 3/4 | 18.0s |
| #21 | GPT-5.4 medium | OpenAI | 1 | 8.3 | 3/4 | 4.11s |
| #24 | GPT-5.2 Chat none | OpenAI | 1 | 8.7 | 3/4 | 3.40s |
| #25 | Qwen3.5 Plus 2026-02-15 medium | Qwen | 1 | 8.2 | 3/4 | 45.8s |
| #28 | Gemini 2.5 Flash medium | 1 | 8.4 | 3/4 | 6.30s | |
| #31 | DeepSeek V4 Flash high | DeepSeek | 1 | 8.3 | 3/4 | 28.5s |
| #45 | GPT-5.4 Mini medium | OpenAI | 1 | 8.6 | 3/4 | 4.05s |
| #48 | Gemini 3 Flash Preview none | 1 | 8.3 | 3/4 | 1.25s | |
| #50 | Gemini 3.1 Flash Lite Preview low | 1 | 8.3 | 3/4 | 2.12s | |
| #53 | Gemini 3.1 Flash Lite high | 1 | 8.7 | 3/4 | 37.2s | |
| #57 | Step 3.7 Flash low | Stepfun | 1 | 8.7 | 3/4 | 4.02s |