AbstentionBench - stale - FreshQA - Recall: leaderboard
AbstentionBench evaluates whether a model appropriately declines unanswerable or underspecified questions. This slice reports stale - FreshQA - Recall as Abstention Recall (%); values converted from 0-1 to percent. Precision and recall must be read together: always abstaining is not necessarily useful. Scenario/dataset pairs and SFT/DPO/RL checkpoints remain separate; no synthetic overall average. Source: facebookresearch/AbstentionBench analysis/abstention_performance.csv.
Metric: Abstention Recall (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B | 89.36 | #499 |
| 2 | OLMo-7B-0724-Instruct-hf | 89.36 | #1441 |
| 3 | Llama-3.1-Tulu-3-8B | 87.23 | #848 |
| 4 | Llama-3.1-Tulu-3-8B-DPO | 85.11 | #992 |
| 5 | Llama-3.1-Tulu-3-70B-DPO | 85.11 | #732 |
| 6 | Qwen 2.5 32B Instruct | 78.72 | #509 |
| 7 | Llama 3.1 8B Instruct | 72.34 | #1017 |
| 8 | Llama 3.1 405B Instruct | 72.34 | #470 |
| 9 | Llama 3.3 70B Instruct | 63.83 | #541 |
| 10 | Llama 3.1 70B Instruct | 59.57 | #567 |
| 11 | Mistral 7B Instruct (v0.3) | 59.57 | #1255 |
| 12 | Llama 3.1 8B | 53.19 | #1130 |
| 13 | Llama-3.1-Tulu-3-8B-SFT | 53.19 | #977 |
| 14 | GPT-4o | 48.94 | #357 |
| 15 | Llama 3.1 70B | 42.55 | #588 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-stale-freshqa-recall · How It Works · Data refreshed daily, snapshot 2026-11-05.