AbstentionBench - underspecified context - UMWP - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-8B | 100 |
| 2 | Llama-3.1-Tulu-3-8B-DPO | 100 |
| 3 | O1 | 99.85 |
| 4 | Llama-3.1-Tulu-3-70B | 99.77 |
| 5 | GPT-4o | 99.74 |
| 6 | Llama-3.1-Tulu-3-70B-DPO | 99.65 |
| 7 | DeepSeek R1 Distill Llama 70B | 99.6 |
| 8 | Llama 3.1 70B Instruct | 99.6 |
| 9 | Llama 3.3 70B Instruct | 99.47 |
| 10 | Llama 3.1 8B Instruct | 99.36 |
| 11 | Qwen 2.5 32B Instruct | 99.14 |
| 12 | Gemini 1.5 Pro | 98.68 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 98.25 |
| 14 | Mistral 7B Instruct (v0.3) | 93.07 |
| 15 | Llama 3.1 70B | 92.12 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-umwp-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.