AbstentionBench - subjective - KUQ/Controversial - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.5 Pro | 75.37 |
| 2 | Llama-3.1-Tulu-3-70B | 71.81 |
| 3 | Llama-3.1-Tulu-3-70B-DPO | 71.51 |
| 4 | GPT-4o | 70.33 |
| 5 | Llama 3.1 8B Instruct | 67.95 |
| 6 | Llama 3.3 70B Instruct | 67.95 |
| 7 | Llama-3.1-Tulu-3-8B-DPO | 64.99 |
| 8 | Llama 3.1 405B Instruct | 64.09 |
| 9 | O1 | 63.2 |
| 10 | Llama-3.1-Tulu-3-8B | 61.42 |
| 11 | Llama 3.1 70B Instruct | 58.46 |
| 12 | Qwen 2.5 32B Instruct | 58.46 |
| 13 | Mistral 7B Instruct (v0.3) | 54.9 |
| 14 | Llama-3.1-Tulu-3-8B-SFT | 51.93 |
| 15 | Llama 3.1 8B | 47.18 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-subjective-kuq-controversial-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.