AbstentionBench - underspecified context - BBQ - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 32B Instruct | 95.78 |
| 2 | Llama 3.1 405B Instruct | 93.04 |
| 3 | Gemini 1.5 Pro | 92.15 |
| 4 | GPT-4o | 92.01 |
| 5 | Llama-3.1-Tulu-3-70B-DPO | 90.84 |
| 6 | Llama-3.1-Tulu-3-70B | 90.84 |
| 7 | Llama 3.1 70B Instruct | 89.96 |
| 8 | Mistral 7B Instruct (v0.3) | 88.46 |
| 9 | Llama 3.3 70B Instruct | 87.94 |
| 10 | Llama 3.1 8B Instruct | 85.05 |
| 11 | DeepSeek R1 Distill Llama 70B | 76.91 |
| 12 | Llama-3.1-Tulu-3-8B-DPO | 76.56 |
| 13 | Llama-3.1-Tulu-3-8B | 74.54 |
| 14 | O1 | 70.17 |
| 15 | Llama 3.1 70B | 61.67 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-bbq-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.