AbstentionBench - false premise - QAQA - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 | 63.86 |
| 2 | Gemini 1.5 Pro | 56.49 |
| 3 | Llama-3.1-Tulu-3-70B | 55.44 |
| 4 | GPT-4o | 54.04 |
| 5 | Llama-3.1-Tulu-3-70B-DPO | 52.28 |
| 6 | Llama 3.1 405B Instruct | 50.18 |
| 7 | Llama-3.1-Tulu-3-8B-DPO | 48.07 |
| 8 | Qwen 2.5 32B Instruct | 46.32 |
| 9 | Mistral 7B Instruct (v0.3) | 44.56 |
| 10 | Llama-3.1-Tulu-3-8B | 42.46 |
| 11 | Llama 3.1 8B Instruct | 41.4 |
| 12 | Llama 3.3 70B Instruct | 41.4 |
| 13 | Llama 3.1 70B Instruct | 40 |
| 14 | Llama-3.1-Tulu-3-8B-SFT | 36.14 |
| 15 | DeepSeek R1 Distill Llama 70B | 31.93 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-false-premise-qaqa-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.