AbstentionBench - subjective - CoCoNot/Humanizing - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-8B-SFT | 98.77 |
| 2 | O1 | 98.14 |
| 3 | Llama-3.1-Tulu-3-8B-DPO | 96.86 |
| 4 | Llama-3.1-Tulu-3-70B-DPO | 96.86 |
| 5 | Llama 3.1 8B Instruct | 96.2 |
| 6 | Llama 3.3 70B Instruct | 96.2 |
| 7 | Llama 3.1 70B Instruct | 95.54 |
| 8 | Llama-3.1-Tulu-3-70B | 95.54 |
| 9 | Llama-3.1-Tulu-3-8B | 94.87 |
| 10 | GPT-4o | 94.19 |
| 11 | Llama 3.1 405B Instruct | 94.19 |
| 12 | Gemini 1.5 Pro | 94.19 |
| 13 | Mistral 7B Instruct (v0.3) | 92.11 |
| 14 | Qwen 2.5 32B Instruct | 90.67 |
| 15 | DeepSeek R1 Distill Llama 70B | 77.61 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-subjective-coconot-humanizing-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.