GLUE MNLI (Matched): leaderboard
Multi-Genre NLI (Williams et al., 2018): label sentence pairs as entailment, neutral or contradiction. This board is accuracy on the matched set, whose genres also appear in the training data, from the archived Papers With Code table of scores as each paper reported them; higher is better.
Metric: Matched. Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 57 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Turing NLR v5 XXL 5.4B (fine-tuned) | 92.6 | |
| 2 | UnitedSynT5 (3B) | 92.6 | |
| 3 | T5 | 92 | |
| 4 | T5-XXL 11B (fine-tuned) | 92 | |
| 5 | T5-3B | 91.4 | |
| 6 | ALBERT | 91.3 | |
| 7 | DeBERTa (large) | 91.1 | |
| 8 | Adv-RoBERTa ensemble | 91.1 | |
| 9 | RoBERTa | 90.8 | |
| 10 | XLNet (single model) | 90.8 |
Interactive version: theaggregate.ai/benchmark?slug=glue-mnli-matched · How It Works · Data refreshed daily, snapshot 2026-11-05.