Stick To Your Role! - Rank-Order Stability: leaderboard
Rank-order stability from Stick To Your Role! (Flowers team, Inria, 2024): the correlation between a simulated persona population's orderings on each Schwartz value in two conversation contexts, averaged over context pairs and values. Range -1 to 1; higher means personas keep their relative values whatever the topic.
Metric: Rank-Order Stability (correlation, -1 to 1). Source: huggingface.co. Saturation forecast: Estimated already saturated. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 0.81 | #845 |
| 2 | Llama 3.3 70B Instruct | 0.81 | #1353 |
| 3 | QwQ-32B | 0.81 | #909 |
| 4 | Llama 3.1 Nemotron 70B Instruct | 0.8 | #1447 |
| 5 | Mistral Large 2 (Jul) | 0.78 | #1153 |
| 6 | Llama 3.1 70B Instruct | 0.76 | #1541 |
| 7 | Qwen 3 32B | 0.74 | #960 |
| 8 | Qwen 3 8B | 0.72 | #1838 |
| 9 | Mistral Large 2 (Nov) Instruct (2411) | 0.71 | #1199 |
| 10 | Mistral Small 3.1 | 0.69 | #1699 |
| 11 | Qwen 3 235B A22B FP8 | 0.68 | #533 |
| 12 | Qwen 3 4B | 0.66 | #2297 |
| 13 | Qwen2.5-14B-Instruct-1M | 0.66 | #1417 |
| 14 | Llama 3.1 8B Instruct | 0.55 | #4277 |
| 15 | GLM-4 32B (0414) | 0.52 | #1407 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=stick-to-your-role-rank-order-stability · How It Works · Data refreshed daily, snapshot 2026-11-12.