AbstentionBench - false premise - FalseQA - F1: leaderboard
Metric: Abstention F1 (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 1.5 Pro | 75.54 |
| 2 | Qwen 2.5 32B Instruct | 74.51 |
| 3 | O1 | 74.26 |
| 4 | GPT-4o | 72.38 |
| 5 | Llama 3.3 70B Instruct | 71.96 |
| 6 | Llama 3.1 405B Instruct | 68.37 |
| 7 | Llama 3.1 70B Instruct | 66.79 |
| 8 | Llama-3.1-Tulu-3-70B-DPO | 64.96 |
| 9 | Mistral 7B Instruct (v0.3) | 64.89 |
| 10 | Llama-3.1-Tulu-3-70B | 64.83 |
| 11 | Llama 3.1 8B Instruct | 60.74 |
| 12 | DeepSeek R1 Distill Llama 70B | 58.74 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 57.17 |
| 14 | Llama-3.1-Tulu-3-8B-DPO | 55.8 |
| 15 | Llama-3.1-Tulu-3-8B | 54.8 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-falseqa-f1 · How It Works · Data refreshed daily, snapshot 2026-09-19.