AbstentionBench - stale - FreshQA - Recall: leaderboard

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

Top models

#ModelScore
1Llama-3.1-Tulu-3-70B89.36
2Llama-3.1-Tulu-3-8B87.23
3Llama-3.1-Tulu-3-8B-DPO85.11
4Llama-3.1-Tulu-3-70B-DPO85.11
5Qwen 2.5 32B Instruct78.72
6Llama 3.1 8B Instruct72.34
7Llama 3.1 405B Instruct72.34
8Llama 3.3 70B Instruct63.83
9Llama 3.1 70B Instruct59.57
10Mistral 7B Instruct (v0.3)59.57
11Llama 3.1 8B53.19
12Llama-3.1-Tulu-3-8B-SFT53.19
13GPT-4o48.94
14Llama 3.1 70B42.55
15Gemini 1.5 Pro36.17

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