AbstentionBench - underspecified intent - SituatedQA/Geo - Recall: leaderboard
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B-DPO | 88.37 |
| 2 | Llama-3.1-Tulu-3-70B | 86.82 |
| 3 | Llama-3.1-Tulu-3-8B-DPO | 79.07 |
| 4 | Llama-3.1-Tulu-3-8B | 76.74 |
| 5 | Qwen 2.5 32B Instruct | 75.97 |
| 6 | O1 | 75.97 |
| 7 | Llama-3.1-Tulu-3-8B-SFT | 73.64 |
| 8 | GPT-4o | 72.09 |
| 9 | Llama 3.1 8B Instruct | 68.22 |
| 10 | Mistral 7B Instruct (v0.3) | 61.24 |
| 11 | Llama 3.3 70B Instruct | 60.47 |
| 12 | Llama 3.1 405B Instruct | 59.69 |
| 13 | Gemini 1.5 Pro | 55.04 |
| 14 | Llama 3.1 70B Instruct | 52.71 |
| 15 | Llama 3.1 70B | 43.41 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-intent-situatedqa-geo-recall · How It Works · Data refreshed daily, snapshot 2026-09-19.