AbstentionBench - answer unknown - KUQ/Unsolved problems - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.5 Pro | 77.45 |
| 2 | Llama-3.1-Tulu-3-70B | 72.34 |
| 3 | GPT-4o | 71.49 |
| 4 | Llama-3.1-Tulu-3-70B-DPO | 70.21 |
| 5 | O1 | 67.66 |
| 6 | Llama-3.1-Tulu-3-8B-DPO | 66.81 |
| 7 | Llama 3.3 70B Instruct | 66.38 |
| 8 | Llama-3.1-Tulu-3-8B | 65.11 |
| 9 | Llama 3.1 8B Instruct | 64.68 |
| 10 | Llama 3.1 405B Instruct | 63.83 |
| 11 | Mistral 7B Instruct (v0.3) | 61.28 |
| 12 | Qwen 2.5 32B Instruct | 60.43 |
| 13 | Llama 3.1 70B Instruct | 58.72 |
| 14 | Llama-3.1-Tulu-3-8B-SFT | 53.19 |
| 15 | Llama 3.1 70B | 51.06 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-kuq-unsolved-problems-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.