AbstentionBench - answer unknown - KUQ/Unsolved problems - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 93.75 |
| 2 | Llama-3.1-Tulu-3-8B | 91.62 |
| 3 | Llama-3.1-Tulu-3-8B-SFT | 91.24 |
| 4 | O1 | 90.86 |
| 5 | Llama-3.1-Tulu-3-8B-DPO | 90.75 |
| 6 | Llama 3.3 70B Instruct | 88.64 |
| 7 | Gemini 1.5 Pro | 88.35 |
| 8 | Llama-3.1-Tulu-3-70B | 88.08 |
| 9 | GPT-4o | 87.96 |
| 10 | Llama-3.1-Tulu-3-70B-DPO | 86.84 |
| 11 | Llama 3.1 70B Instruct | 86.25 |
| 12 | Qwen 2.5 32B Instruct | 86.06 |
| 13 | Mistral 7B Instruct (v0.3) | 85.71 |
| 14 | Llama 3.1 70B | 84.51 |
| 15 | Llama 3.1 8B | 83.94 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-kuq-unsolved-problems-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.