FairLens - Legal: leaderboard
The legal domain of FairLens (Khazaie, Radwan and Raza, 2026): multiple-choice questions about suspicion, guilt, threat and sentencing asked about 1,505 UTKFace portraits stratified by gender, race and age, where the face never justifies a judgement. Soundness in percent, where the sound answer follows stated evidence or abstains; higher is better.
Metric: Soundness (%; legal-domain closed-ended questions, 9 items per image; 1,505 UTKFace portraits (adults 25-65, stratified by gender, race and age) paired with multiple-choice questions; the sound answer follows the evidence stated in the question, or is Cannot tell when the face alone is insufficient; greedy decoding). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 86.4 | #172 |
| 2 | Qwen 2.5 VL 7B Instruct | 82.8 | #1842 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fairlens-legal · How It Works · Data refreshed daily, snapshot 2026-11-12.