AbstentionBench - stale - FreshQA - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.1 70B | 62.5 |
| 2 | O1 | 53.13 |
| 3 | Gemini 1.5 Pro | 51.52 |
| 4 | Llama 3.1 8B | 51.02 |
| 5 | GPT-4o | 46 |
| 6 | DeepSeek R1 Distill Llama 70B | 45.45 |
| 7 | Llama-3.1-Tulu-3-8B | 36.61 |
| 8 | Llama 3.3 70B Instruct | 35.71 |
| 9 | Llama-3.1-Tulu-3-8B-SFT | 34.72 |
| 10 | Llama-3.1-Tulu-3-8B-DPO | 34.19 |
| 11 | Llama-3.1-Tulu-3-70B-DPO | 33.9 |
| 12 | Llama-3.1-Tulu-3-70B | 33.33 |
| 13 | Qwen 2.5 32B Instruct | 31.9 |
| 14 | Llama 3.1 70B Instruct | 30.11 |
| 15 | Llama 3.1 405B Instruct | 29.82 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-stale-freshqa-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.