AbstentionBench - underspecified context - GPQA-Diamond - Recall: leaderboard

Metric: Abstention Recall (%). Source: github.com. 21 models tracked.

Top models

#ModelScore
1O177.5
2GPT-4o75
3Llama 3.1 405B Instruct65
4Qwen 2.5 32B Instruct65
5Llama-3.1-Tulu-3-70B65
6Llama-3.1-Tulu-3-70B-DPO65
7O1 (Low)62.5
8O1 (High)60
9Llama 3.1 8B Instruct52.5
10Mistral 7B Instruct (v0.3)52.5
11Gemini 1.5 Pro50
12Llama 3.1 70B Instruct45
13Llama 3.3 70B Instruct37.5
14Llama-3.1-Tulu-3-8B-SFT32.5
15DeepSeek R1 Distill Llama 70B25

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