AbstentionBench - stale - FreshQA - Precision: leaderboard

Metric: Abstention Precision (%). Source: github.com. 20 models tracked.

Top models

#ModelScore
1Llama 3.1 70B62.5
2O153.13
3Gemini 1.5 Pro51.52
4Llama 3.1 8B51.02
5GPT-4o46
6DeepSeek R1 Distill Llama 70B45.45
7Llama-3.1-Tulu-3-8B36.61
8Llama 3.3 70B Instruct35.71
9Llama-3.1-Tulu-3-8B-SFT34.72
10Llama-3.1-Tulu-3-8B-DPO34.19
11Llama-3.1-Tulu-3-70B-DPO33.9
12Llama-3.1-Tulu-3-70B33.33
13Qwen 2.5 32B Instruct31.9
14Llama 3.1 70B Instruct30.11
15Llama 3.1 405B Instruct29.82

Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-stale-freshqa-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.