RACE (Middle): leaderboard
RACE (Lai et al., 2017): multiple-choice reading comprehension from English exams for Chinese students aged 12 to 18. This board is accuracy on the middle-school subset (RACE-M), from the archived Papers With Code table of scores as each paper reported them, under that paper’s own setup; higher is better.
Metric: Accuracy (Middle). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Megatron-BERT (ensemble) | 93.1 | |
| 2 | Megatron-BERT | 91.8 | |
| 3 | B10-10-10 | 88.8 | |
| 4 | ALBERTxxlarge+DUMA(ensemble) | 88.7 | |
| 5 | XLNet | 88.6 | |
| 6 | RoBERTa | 86.5 | |
| 7 | PaLM 540B (zero-shot) | 68.1 | |
| 8 | LLaMA 65B (zero-shot) | 67.9 | |
| 9 | PaLM 62B (zero-shot) | 64.3 | |
| 10 | LLaMA 33B (zero-shot) | 64.1 |
Interactive version: theaggregate.ai/benchmark?slug=race-middle · How It Works · Data refreshed daily, snapshot 2026-11-05.