AbstentionBench - underspecified context - UMWP - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 86.38 |
| 2 | Llama 3.1 70B Instruct | 82.65 |
| 3 | Gemini 1.5 Pro | 80.68 |
| 4 | GPT-4o | 79.56 |
| 5 | Qwen 2.5 32B Instruct | 79.41 |
| 6 | Llama 3.1 8B Instruct | 76.09 |
| 7 | Mistral 7B Instruct (v0.3) | 73.43 |
| 8 | DeepSeek R1 Distill Llama 70B | 72.25 |
| 9 | Llama 3.3 70B Instruct | 69.47 |
| 10 | Llama 3.1 70B | 68.2 |
| 11 | Llama-3.1-Tulu-3-70B | 66.46 |
| 12 | Llama-3.1-Tulu-3-70B-DPO | 65.82 |
| 13 | Llama 3.1 8B | 64.46 |
| 14 | Llama-3.1-Tulu-3-8B-DPO | 48.22 |
| 15 | Llama-3.1-Tulu-3-8B | 48.02 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-umwp-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.