SLM Trustworthiness - Fairness: leaderboard
SLM Trustworthiness (Benchmarking Trustworthiness of SLMs, arXiv 2608.11981v1, Tables I, II and IV): TrustLLM evaluation of uncompressed instruction-tuned models from 360M to 8B parameters, the baselines of a pruning, quantization and distillation study. Fairness: mean score over disparagement, representation bias and preference bias in subjective choices, percent; it does not rise with model size here. Higher is better.
Metric: TrustLLM fairness accuracy (%; sub-task mean). Source: arxiv.org. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 1.5B Instruct | 71.82 | #1393 |
| 2 | Llama 3.2 1B Instruct | 70.81 | #1518 |
| 3 | Qwen 2.5 0.5B Instruct | 48.99 | #1510 |
| 4 | SmolLM2-360M-Instruct | 47.45 | #1594 |
| 5 | Gemma 1.1 7B (IT) | 44.73 | #1348 |
| 6 | Llama 3.2 3B Instruct | 43.3 | #1327 |
| 7 | h2o-danube3-500m-chat | 43.27 | #1574 |
| 8 | Llama 3.1 8B Instruct | 36.03 | #1017 |
| 9 | Qwen 2.5 3B Instruct | 29.99 | #1159 |
| 10 | Qwen 2.5 7B Instruct | 28.33 | #863 |
Interactive version: theaggregate.ai/benchmark?slug=slm-trustworthiness-fairness · How It Works · Data refreshed daily, snapshot 2026-11-05.