AbstentionBench - false premise - KUQ/False assumptions - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Llama 70B | 96.18 |
| 2 | O1 | 94.81 |
| 3 | Llama-3.1-Tulu-3-8B-SFT | 93.23 |
| 4 | Gemini 1.5 Pro | 92.31 |
| 5 | Llama-3.1-Tulu-3-8B-DPO | 91.93 |
| 6 | Llama-3.1-Tulu-3-8B | 91.52 |
| 7 | GPT-4o | 91.22 |
| 8 | Llama 3.1 8B | 90.98 |
| 9 | Llama 3.3 70B Instruct | 90.5 |
| 10 | Llama-3.1-Tulu-3-70B | 90.26 |
| 11 | Llama-3.1-Tulu-3-70B-DPO | 90.21 |
| 12 | Llama 3.1 70B | 89.13 |
| 13 | Mistral 7B Instruct (v0.3) | 87.82 |
| 14 | Qwen 2.5 32B Instruct | 86.91 |
| 15 | Llama 3.1 70B Instruct | 86.44 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-kuq-false-assumptions-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.