AbstentionBench - underspecified context - BB/Disambiguate - Precision: leaderboard
AbstentionBench evaluates whether a model appropriately declines unanswerable or underspecified questions. This slice reports underspecified context - BB/Disambiguate - Precision as Abstention Precision (%); 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 Precision (%). Source: github.com. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B | 72.73 | #499 |
| 2 | Llama-3.1-Tulu-3-70B-DPO | 63.41 | #732 |
| 3 | DeepSeek R1 Distill Llama 70B | 54.05 | #614 |
| 4 | Llama 3.1 8B | 53.57 | #1130 |
| 5 | Llama-3.1-Tulu-3-8B | 53.57 | #848 |
| 6 | Llama 3.1 70B | 52.67 | #588 |
| 7 | OLMo-7B-0724-Instruct-hf | 52.38 | #1441 |
| 8 | Llama 3.1 405B Instruct | 48.53 | #470 |
| 9 | Mistral 7B Instruct (v0.3) | 47.62 | #1255 |
| 10 | O1 | 46.67 | #190 |
| 11 | Qwen 2.5 32B Instruct | 46.53 | #509 |
| 12 | Llama-3.1-Tulu-3-8B-DPO | 46.15 | #992 |
| 13 | Llama 3.1 70B Instruct | 45 | #567 |
| 14 | Llama 3.3 70B Instruct | 43.98 | #541 |
| 15 | Gemini 1.5 Pro | 39.74 | #454 |
Interactive version: theaggregate.ai/benchmark?slug=abstentionbench-underspecified-context-bb-disambiguate-precision · How It Works · Data refreshed daily, snapshot 2026-11-05.