AbstentionBench - answer unknown - BB/Known unknowns - Precision: leaderboard
Metric: Abstention Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3.3 70B Instruct | 100 |
| 2 | GPT-4o | 100 |
| 3 | Llama 3.1 8B | 100 |
| 4 | Qwen 2.5 32B Instruct | 100 |
| 5 | Llama 3.1 70B Instruct | 95.83 |
| 6 | Gemini 1.5 Pro | 95.83 |
| 7 | O1 | 95.83 |
| 8 | Llama-3.1-Tulu-3-70B | 95.83 |
| 9 | Llama-3.1-Tulu-3-8B | 95.65 |
| 10 | Llama-3.1-Tulu-3-8B-SFT | 95.45 |
| 11 | DeepSeek R1 Distill Llama 70B | 93.33 |
| 12 | Llama 3.1 8B Instruct | 92 |
| 13 | Llama 3.1 405B Instruct | 92 |
| 14 | Llama-3.1-Tulu-3-70B-DPO | 92 |
| 15 | Mistral 7B Instruct (v0.3) | 85.19 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-answer-unknown-bb-known-unknowns-precision · How It Works · Data refreshed daily, snapshot 2026-09-19.