AbstentionBench - underspecified context - BBQ - Precision: leaderboard

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

Top models

#ModelScore
1DeepSeek R1 Distill Llama 70B98.59
2Llama 3.3 70B Instruct98
3O197.6
4Llama 3.1 70B Instruct96.58
5Qwen 2.5 32B Instruct96.52
6Llama 3.1 405B Instruct96.5
7GPT-4o94.07
8Llama-3.1-Tulu-3-70B-DPO94.04
9Llama-3.1-Tulu-3-70B93.77
10Llama 3.1 70B93.28
11Llama-3.1-Tulu-3-8B93.24
12Llama 3.1 8B91.95
13Gemini 1.5 Pro91.89
14Llama-3.1-Tulu-3-8B-DPO91.41
15Mistral 7B Instruct (v0.3)86.26

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