AbstentionBench - underspecified context - WorldSense - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 98.96 |
| 2 | DeepSeek R1 Distill Llama 70B | 97.45 |
| 3 | Llama 3.1 70B | 94.21 |
| 4 | Llama 3.1 8B | 90.51 |
| 5 | Llama 3.3 70B Instruct | 84.14 |
| 6 | Llama-3.1-Tulu-3-70B | 83.8 |
| 7 | Mistral 7B Instruct (v0.3) | 83.33 |
| 8 | Llama-3.1-Tulu-3-70B-DPO | 81.25 |
| 9 | Llama-3.1-Tulu-3-8B-DPO | 76.27 |
| 10 | Qwen 2.5 32B Instruct | 74.77 |
| 11 | Llama-3.1-Tulu-3-8B | 72.22 |
| 12 | GPT-4o | 67.94 |
| 13 | Llama 3.1 70B Instruct | 67.48 |
| 14 | Llama 3.1 405B Instruct | 67.13 |
| 15 | Llama-3.1-Tulu-3-8B-SFT | 62.85 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-worldsense-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.