AbstentionBench - false premise - FalseQA - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 93.63 |
| 2 | GPT-4o | 90.59 |
| 3 | Qwen 2.5 32B Instruct | 89.07 |
| 4 | Llama 3.1 70B Instruct | 88.67 |
| 5 | Llama 3.3 70B Instruct | 88.35 |
| 6 | Llama-3.1-Tulu-3-70B | 88.22 |
| 7 | Mistral 7B Instruct (v0.3) | 87.62 |
| 8 | Llama-3.1-Tulu-3-8B | 87.54 |
| 9 | Llama-3.1-Tulu-3-70B-DPO | 87.44 |
| 10 | Llama 3.1 405B Instruct | 86.25 |
| 11 | Llama-3.1-Tulu-3-8B-DPO | 85.8 |
| 12 | O1 | 84.99 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 84.38 |
| 14 | Llama 3.1 8B Instruct | 83.46 |
| 15 | Gemini 1.5 Pro | 83.45 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-falseqa-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.