AbstentionBench - underspecified context - SQuAD 2.0 - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 32B Instruct | 84.13 |
| 2 | Llama 3.1 405B Instruct | 83.56 |
| 3 | Llama 3.3 70B Instruct | 81.24 |
| 4 | O1 | 80.86 |
| 5 | Llama 3.1 70B Instruct | 76.91 |
| 6 | Gemini 1.5 Pro | 76.77 |
| 7 | Mistral 7B Instruct (v0.3) | 74.54 |
| 8 | Llama-3.1-Tulu-3-70B-DPO | 73.56 |
| 9 | GPT-4o | 72.79 |
| 10 | Llama 3.1 8B Instruct | 72.52 |
| 11 | Llama-3.1-Tulu-3-70B | 72.22 |
| 12 | Llama 3.1 70B | 70.9 |
| 13 | Llama 3.1 8B | 68.04 |
| 14 | DeepSeek R1 Distill Llama 70B | 64.82 |
| 15 | Llama-3.1-Tulu-3-8B-DPO | 46.36 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-squad-2-0-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.