SQuAD 1.1 (EM): leaderboard
SQuAD 1.1 (Rajpurkar et al., 2016): extractive reading comprehension on Wikipedia passages, where every question is answerable by a span of the passage. This board is exact match, the share of answers matching a reference answer exactly, from the archived Papers With Code table of reported scores; higher is better.
Metric: EM. Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 196 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | {ANNA} (single model) | 90.62 | |
| 2 | LUKE (single model) | 90.2 | |
| 3 | LUKE | 90.2 | |
| 4 | XLNet (single model) | 89.9 | |
| 5 | XLNET-123++ (single model) | 89.86 | |
| 6 | XLNET-123+ (single model) | 89.71 | |
| 7 | XLNET-123 (single model) | 89.65 | |
| 8 | Unnamed submission by NMC | 88.91 | |
| 9 | BERTSP (single model) | 88.91 | |
| 10 | SpanBERT (single model) | 88.84 |
Interactive version: theaggregate.ai/benchmark?slug=squad-1-1-em · How It Works · Data refreshed daily, snapshot 2026-11-05.