AbstentionBench - underspecified context - QASPER - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-8B-DPO | 29.07 |
| 2 | Llama-3.1-Tulu-3-8B | 28.79 |
| 3 | Llama-3.1-Tulu-3-70B-DPO | 26.64 |
| 4 | DeepSeek R1 Distill Llama 70B | 26.47 |
| 5 | Qwen 2.5 32B Instruct | 26.28 |
| 6 | Llama 3.3 70B Instruct | 26.03 |
| 7 | Llama 3.1 8B Instruct | 25.99 |
| 8 | Llama-3.1-Tulu-3-70B | 25.43 |
| 9 | Mistral 7B Instruct (v0.3) | 25.08 |
| 10 | Llama 3.1 405B Instruct | 25 |
| 11 | GPT-4o | 23.68 |
| 12 | Llama 3.1 70B Instruct | 23.59 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 23.14 |
| 14 | Gemini 1.5 Pro | 21.49 |
| 15 | Llama 3.1 70B | 20.91 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-qasper-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.