AbstentionBench - false premise - CoCoNot/False presumptions - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 89.57 |
| 2 | Llama-3.1-Tulu-3-70B | 88.16 |
| 3 | Llama-3.1-Tulu-3-70B-DPO | 86.84 |
| 4 | GPT-4o | 84.35 |
| 5 | Llama 3.1 405B Instruct | 83.54 |
| 6 | Qwen 2.5 32B Instruct | 83.22 |
| 7 | Llama 3.3 70B Instruct | 82.99 |
| 8 | Llama-3.1-Tulu-3-8B-DPO | 82.89 |
| 9 | Llama-3.1-Tulu-3-8B | 81.63 |
| 10 | Mistral 7B Instruct (v0.3) | 80.82 |
| 11 | Llama 3.1 70B Instruct | 80 |
| 12 | DeepSeek R1 Distill Llama 70B | 79.14 |
| 13 | Gemini 1.5 Pro | 78.91 |
| 14 | Llama-3.1-Tulu-3-8B-SFT | 77.37 |
| 15 | Llama 3.1 8B Instruct | 76.51 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-coconot-false-presumptions-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.