StickToYourRole: leaderboard
Role-playing consistency benchmark: measures how stably LLMs maintain simulated personal values during multi-turn role-playing conversations using Schwartz value theory.
Metric: Cardinal Score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 83.6 | #377 |
| 2 | Llama 3.1 Nemotron 70B Instruct | 80.66 | #554 |
| 3 | Mistral Large 2 (Jul) | 78.65 | #484 |
| 4 | Llama 3.3 70B Instruct | 78.26 | #534 |
| 5 | QwQ-32B | 77.19 | #421 |
| 6 | Llama 3.1 70B Instruct | 77.16 | #561 |
| 7 | Qwen 3 32B | 73.64 | #443 |
| 8 | Mistral Large 2 (Nov) Instruct (2411) | 73.35 | #465 |
| 9 | Qwen 3 8B | 71.84 | #674 |
| 10 | Qwen 3 235B A22B FP8 | 71.84 | #301 |
| 11 | Mistral Small 3.1 | 70.26 | #610 |
| 12 | Qwen2.5-14B-Instruct-1M | 70.22 | #578 |
| 13 | Qwen 3 4B | 69.71 | #829 |
| 14 | Llama 3.1 8B Instruct | 62 | #1035 |
| 15 | Llama 4 Scout Instruct | 61.81 | #539 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sticktoyourrole · How It Works · Data refreshed daily, snapshot 2026-10-20.