AbstentionBench - answer unknown - KUQ/Future unknowns - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 97.25 |
| 2 | Llama-3.1-Tulu-3-8B | 94.47 |
| 3 | O1 | 94.35 |
| 4 | Llama 3.1 8B | 93.33 |
| 5 | Llama 3.3 70B Instruct | 92.31 |
| 6 | Llama-3.1-Tulu-3-8B-SFT | 92.31 |
| 7 | Llama-3.1-Tulu-3-8B-DPO | 91.95 |
| 8 | Llama-3.1-Tulu-3-70B | 91.26 |
| 9 | Llama 3.1 70B | 90.81 |
| 10 | GPT-4o | 90.63 |
| 11 | Llama-3.1-Tulu-3-70B-DPO | 89.26 |
| 12 | Gemini 1.5 Pro | 89.17 |
| 13 | Qwen 2.5 32B Instruct | 88.93 |
| 14 | Llama 3.1 70B Instruct | 88.65 |
| 15 | Mistral 7B Instruct (v0.3) | 86.43 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-kuq-future-unknowns-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.