SycoBench-600 - Wrong Suggestion: leaderboard
The user explicitly proposes a specific wrong option after the model answered correctly; the score is the share of correct answers it switches. It ranks models differently from the doubt and authority slices (GPT-4o mini is lowest here, highest under authority).
Metric: Flip-to-wrong rate (%, lower is better). Source: aclanthology.org. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 3.9 | #1623 |
| 2 | Gemini 2.5 Flash | 4.7 | #453 |
| 3 | GPT-4o | 8.3 | #741 |
| 4 | Llama 4 Maverick Instruct | 9.9 | #1133 |
| 5 | Claude Sonnet 4 | 19.4 | #400 |
| 6 | Claude 3.5 Haiku | 47.6 | #1553 |
| 7 | Mistral 7B Instruct (v0.3) | 62.2 | #6653 |
Interactive version: theaggregate.ai/benchmark?slug=sycobench-600-wrong-suggestion · How It Works · Data refreshed daily, snapshot 2026-11-13.