SNLI: leaderboard
Stanford Natural Language Inference (Bowman et al., 2015): about 570,000 human-written English sentence pairs built on image captions, labelled entailment, contradiction or neutral. Archived Papers With Code table of test accuracies as each paper reported them; higher is better.
Metric: % Test Accuracy. Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 90 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | UnitedSynT5 (3B) | 94.7 | |
| 2 | UnitedSynT5 (335M) | 93.5 | |
| 3 | Neural Tree Indexers for Text Understanding | 93.1 | |
| 4 | EFL (Entailment as Few-shot Learner) + RoBERTa-large | 93.1 | |
| 5 | RoBERTa-large+Self-Explaining | 92.3 | |
| 6 | RoBERTa-large + self-explaining layer | 92.3 | |
| 7 | CA-MTL | 92.1 | |
| 8 | SemBERT | 91.9 | |
| 9 | MT-DNN-SMARTLARGEv0 | 91.7 | |
| 10 | MT-DNN | 91.6 |
Interactive version: theaggregate.ai/benchmark?slug=snli · How It Works · Data refreshed daily, snapshot 2026-11-05.