AbstentionBench - underspecified intent - KUQ/Ambiguous - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 82.63 |
| 2 | Llama 3.1 70B | 79.78 |
| 3 | Llama-3.1-Tulu-3-8B-SFT | 79.33 |
| 4 | Llama-3.1-Tulu-3-8B | 79.22 |
| 5 | O1 | 79.05 |
| 6 | GPT-4o | 77.63 |
| 7 | Llama-3.1-Tulu-3-8B-DPO | 76.73 |
| 8 | Llama 3.1 8B | 76.56 |
| 9 | Gemini 1.5 Pro | 74.27 |
| 10 | Llama 3.3 70B Instruct | 73.88 |
| 11 | Llama-3.1-Tulu-3-70B | 70.89 |
| 12 | Llama-3.1-Tulu-3-70B-DPO | 70.71 |
| 13 | Mistral 7B Instruct (v0.3) | 70.63 |
| 14 | Qwen 2.5 32B Instruct | 69.18 |
| 15 | Llama 3.1 70B Instruct | 67.49 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-intent-kuq-ambiguous-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.