AbstentionBench - answer unknown - CoCoNot/Unsupported - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B-DPO | 97.46 |
| 2 | Llama-3.1-Tulu-3-70B | 96.23 |
| 3 | Llama-3.1-Tulu-3-8B-SFT | 95.87 |
| 4 | Gemini 1.5 Pro | 95.76 |
| 5 | Qwen 2.5 32B Instruct | 95.69 |
| 6 | Llama-3.1-Tulu-3-8B-DPO | 95.69 |
| 7 | GPT-4o | 95.24 |
| 8 | Llama-3.1-Tulu-3-8B | 94.83 |
| 9 | Llama 3.1 405B Instruct | 92.51 |
| 10 | O1 | 92.38 |
| 11 | Llama 3.1 8B Instruct | 92.04 |
| 12 | Llama 3.3 70B Instruct | 90.75 |
| 13 | Llama 3.1 70B Instruct | 90.58 |
| 14 | Mistral 7B Instruct (v0.3) | 89.91 |
| 15 | DeepSeek R1 Distill Llama 70B | 78.17 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-coconot-unsupported-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.