AbstentionBench - answer unknown - KUQ/Future unknowns - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B | 85.29 |
| 2 | Llama-3.1-Tulu-3-70B-DPO | 85.26 |
| 3 | Llama-3.1-Tulu-3-8B | 82.56 |
| 4 | Gemini 1.5 Pro | 81.92 |
| 5 | Llama-3.1-Tulu-3-8B-DPO | 81.77 |
| 6 | GPT-4o | 79.73 |
| 7 | Llama 3.3 70B Instruct | 79.58 |
| 8 | O1 | 78.06 |
| 9 | Llama 3.1 8B Instruct | 76.77 |
| 10 | Mistral 7B Instruct (v0.3) | 76.37 |
| 11 | Qwen 2.5 32B Instruct | 76.14 |
| 12 | Llama-3.1-Tulu-3-8B-SFT | 74.59 |
| 13 | Llama 3.1 405B Instruct | 74.35 |
| 14 | Llama 3.1 70B Instruct | 73.15 |
| 15 | DeepSeek R1 Distill Llama 70B | 69.69 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-kuq-future-unknowns-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.