AbstentionBench - underspecified context - QASPER - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.5 Pro | 98.73 |
| 2 | Qwen 2.5 32B Instruct | 97.47 |
| 3 | Llama-3.1-Tulu-3-70B-DPO | 97.47 |
| 4 | Llama 3.3 70B Instruct | 96.2 |
| 5 | GPT-4o | 96.2 |
| 6 | Llama 3.1 405B Instruct | 96.2 |
| 7 | O1 | 96.2 |
| 8 | Llama-3.1-Tulu-3-8B | 96.2 |
| 9 | Llama-3.1-Tulu-3-8B-DPO | 94.94 |
| 10 | Mistral 7B Instruct (v0.3) | 93.67 |
| 11 | Llama-3.1-Tulu-3-70B | 93.67 |
| 12 | Llama 3.1 8B Instruct | 91.14 |
| 13 | DeepSeek R1 Distill Llama 70B | 91.14 |
| 14 | Llama 3.1 70B Instruct | 89.87 |
| 15 | Llama 3.1 70B | 87.34 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-qasper-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.