AbstentionBench - underspecified context - ALCUNA - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 32B Instruct | 83.25 |
| 2 | Llama-3.1-Tulu-3-70B-DPO | 81.87 |
| 3 | Llama-3.1-Tulu-3-70B | 81.75 |
| 4 | Gemini 1.5 Pro | 81.39 |
| 5 | Llama 3.1 8B Instruct | 81.04 |
| 6 | Llama 3.1 405B Instruct | 80.26 |
| 7 | Llama 3.3 70B Instruct | 79.91 |
| 8 | GPT-4o | 79.64 |
| 9 | Llama 3.1 70B Instruct | 79.41 |
| 10 | Mistral 7B Instruct (v0.3) | 78.41 |
| 11 | DeepSeek R1 Distill Llama 70B | 74.63 |
| 12 | Llama 3.1 70B | 72.47 |
| 13 | O1 | 65.62 |
| 14 | Llama-3.1-Tulu-3-8B-DPO | 64.4 |
| 15 | Llama-3.1-Tulu-3-8B | 62.86 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-alcuna-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.