The Wrong Kind of Right: leaderboard
Warning: This paper studies stereotypes and biases, and contains potentially disturbing examples, used for illustration purposes only. Our findings should not be interpreted as an argument against alignment. Instead, this paper highlights the need for principled approaches to more advanced alignment. Alignment aims to ensure that large language models (LLMs) behave safely and reliably, including b
Metric: Misfired Alignment Rate, average over 8 categories (%) (self-reported). Source: benchmarklist.com. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 | 4.7 | #251 |
| 2 | Gemini 3.1 Pro (Preview) | 4.9 | #58 |
| 3 | DeepSeek V3 Chat | 5.2 | #447 |
| 4 | Qwen 3.5 9B | 5.5 | #379 |
| 5 | Qwen 3.5 27B | 5.5 | #187 |
| 6 | Mistral 7B | 5.7 | #1397 |
| 7 | Llama 3.1 8B | 6.2 | #1151 |
| 8 | Gemma 3 27B | 6.3 | #607 |
| 9 | GPT-5.5 | 7.5 | #31 |
| 10 | Qwen 3.5 4B | 7.9 | #475 |
| 11 | Qwen 3 4B | 7.9 | #829 |
| 12 | Grok 4.20 | 7.9 | #175 |
| 13 | Gemini 3.1 Flash Lite | 8.3 | #191 |
| 14 | Llama 3.1 70B | 8.6 | #592 |
| 15 | Qwen 3 32B | 8.7 | #443 |
Interactive version: theaggregate.ai/benchmark?slug=the-wrong-kind-of-right · How It Works · Data refreshed daily, snapshot 2026-10-20.