WinoGrande (PapersWithCode): leaderboard
WinoGrande (Sakaguchi et al., 2020): about 44,000 two-option fill-in-the-blank problems in the style of the Winograd Schema Challenge, filtered to reduce dataset biases. This board is the archived Papers With Code table of accuracies as each paper reported them, under that paper’s own setup, not a common evaluation run; higher is better.
Metric: Accuracy. Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 76 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | ST-MoE-32B 269B (fine-tuned) | 96.1 | |
| 2 | Unicorn 11B (fine-tuned) | 91.3 | |
| 3 | CompassMTL 567M with Tailor | 90.5 | |
| 4 | CompassMTL 567M | 89.6 | |
| 5 | UnifiedQA 11B (fine-tuned) | 89.4 | |
| 6 | Claude 3 Opus (5-shot) | 88.5 | |
| 7 | GPT-4 (5-shot) | 87.5 | |
| 8 | ExDeBERTa 567M | 87 | |
| 9 | LLaMA-2 13B + MixLoRA | 86.3 | |
| 10 | LLaMA3 8B+MoSLoRA | 85.8 |
Interactive version: theaggregate.ai/benchmark?slug=winogrande-paperswithcode · How It Works · Data refreshed daily, snapshot 2026-11-05.