| Char-level CNN+LSTM (partial scoring) | 57.9 | A Simple Method for Commonsense Reasoning | |
| BERT-base 110M (fine-tuned on WSCR) | 62.3 | A Surprisingly Robust Trick for Winograd Schema Challenge | |
| BERTwiki 340M (fine-tuned on WSCR) | 72.5 | A Surprisingly Robust Trick for Winograd Schema Challenge | |
| USSM + Supervised DeepNet + KB | 52.8 | Attention Is (not) All You Need for Commonsense Reasoning | |
| Hybrid H3 125M (3-shot, logit scoring) | 43.3 | Hungry Hungry Hippos: Towards Language Modeling with State Space Models | |
| GPT-3 175B (few-shot) | 80.1 | Language Models are Few-Shot Learners | |