POLAR-Bench - Utility: leaderboard
Metric: Utility score (%): share of the task-relevant attributes the model shared to complete the task, trusted agent holding a source document, a privacy policy and a task, questioned single-turn or multi-turn by a fixed Llama-3.3-70B-Instruct third party under five attack strategies, attributes revealed in the transcript extracted by normalization and regex, mean over the 10 domains; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GLM-5.1 | 89.46 |
| 2 | Gemma 4 E2B | 87.56 |
| 3 | Gemma 4 E4B | 86.29 |
| 4 | GPT-5.4 | 84.79 |
| 5 | Gemma 3 27B | 84.71 |
| 6 | Ministral 3 14B | 84.48 |
| 7 | Qwen 3 32B | 84.43 |
| 8 | Ministral 3 8B | 83.7 |
| 9 | DeepSeek R1 Distill Llama 70B | 82.93 |
| 10 | Gemma 4 31B | 82.77 |
| 11 | DeepSeek V3.1 | 82.44 |
| 12 | Apertus-70B-Instruct-2509 | 81.8 |
| 13 | Ministral 3 3B | 81.46 |
| 14 | SmolLM3-3B | 77.66 |
| 15 | DeepSeek R1 Distill Qwen 32B | 77.6 |
Interactive version: theaggregate.ai/benchmark?slug=polar-bench-utility · How It Works · Data refreshed daily, snapshot 2026-10-07.