AbstentionBench - subjective - CoCoNot/Humanizing - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-8B-SFT | 97.56 |
| 2 | O1 | 96.34 |
| 3 | Llama-3.1-Tulu-3-8B-DPO | 93.9 |
| 4 | Llama-3.1-Tulu-3-70B-DPO | 93.9 |
| 5 | Llama 3.1 8B Instruct | 92.68 |
| 6 | Llama 3.3 70B Instruct | 92.68 |
| 7 | Llama 3.1 70B Instruct | 91.46 |
| 8 | Llama-3.1-Tulu-3-70B | 91.46 |
| 9 | Llama-3.1-Tulu-3-8B | 90.24 |
| 10 | GPT-4o | 89.02 |
| 11 | Llama 3.1 405B Instruct | 89.02 |
| 12 | Gemini 1.5 Pro | 89.02 |
| 13 | Mistral 7B Instruct (v0.3) | 85.37 |
| 14 | Qwen 2.5 32B Instruct | 82.93 |
| 15 | DeepSeek R1 Distill Llama 70B | 63.41 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-subjective-coconot-humanizing-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.