AbstentionBench - underspecified context - SQuAD 2.0 - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 32B Instruct | 74.99 |
| 2 | Llama 3.1 405B Instruct | 73.35 |
| 3 | Llama 3.3 70B Instruct | 69.68 |
| 4 | O1 | 69.17 |
| 5 | Llama 3.1 70B Instruct | 63.47 |
| 6 | Gemini 1.5 Pro | 62.9 |
| 7 | Mistral 7B Instruct (v0.3) | 61.26 |
| 8 | Llama-3.1-Tulu-3-70B-DPO | 59.46 |
| 9 | Llama 3.1 8B Instruct | 58.05 |
| 10 | GPT-4o | 57.76 |
| 11 | Llama-3.1-Tulu-3-70B | 57.54 |
| 12 | Llama 3.1 70B | 55.11 |
| 13 | Llama 3.1 8B | 52.17 |
| 14 | DeepSeek R1 Distill Llama 70B | 48.33 |
| 15 | Llama-3.1-Tulu-3-8B-DPO | 30.38 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-squad-2-0-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.