AbstentionBench - underspecified context - WorldSense - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 93.85 |
| 2 | DeepSeek R1 Distill Llama 70B | 91.47 |
| 3 | Llama 3.1 70B | 84.66 |
| 4 | Llama 3.1 8B | 77.5 |
| 5 | Qwen 2.5 32B Instruct | 76 |
| 6 | Llama 3.1 405B Instruct | 75.97 |
| 7 | GPT-4o | 75.45 |
| 8 | Llama 3.3 70B Instruct | 73.47 |
| 9 | Llama 3.1 70B Instruct | 67.09 |
| 10 | Llama-3.1-Tulu-3-70B | 63.99 |
| 11 | Llama-3.1-Tulu-3-70B-DPO | 62.79 |
| 12 | Gemini 1.5 Pro | 60.63 |
| 13 | Mistral 7B Instruct (v0.3) | 54.88 |
| 14 | Llama-3.1-Tulu-3-8B-DPO | 51.73 |
| 15 | Llama-3.1-Tulu-3-8B | 51.23 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-worldsense-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.