AbstentionBench - false premise - CoCoNot/False presumptions - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 87.95 |
| 2 | Llama-3.1-Tulu-3-70B | 80.72 |
| 3 | Llama 3.1 405B Instruct | 79.52 |
| 4 | Llama-3.1-Tulu-3-70B-DPO | 79.52 |
| 5 | Llama-3.1-Tulu-3-8B-DPO | 75.9 |
| 6 | GPT-4o | 74.7 |
| 7 | Qwen 2.5 32B Instruct | 74.7 |
| 8 | Llama 3.3 70B Instruct | 73.49 |
| 9 | Llama-3.1-Tulu-3-8B | 72.29 |
| 10 | Mistral 7B Instruct (v0.3) | 71.08 |
| 11 | Llama 3.1 70B Instruct | 69.88 |
| 12 | Gemini 1.5 Pro | 69.88 |
| 13 | Llama 3.1 8B Instruct | 68.67 |
| 14 | DeepSeek R1 Distill Llama 70B | 66.27 |
| 15 | Llama-3.1-Tulu-3-8B-SFT | 63.86 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-coconot-false-presumptions-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.