Best Accuracy
Updated Daily
Google's PaLM 540B achieves breakthrough accuracy across reasoning, language understanding, and generation tasks.
Google's GLaM achieves high accuracy with a sparse mixture-of-experts architecture, surpassing dense models on several benchmarks.
DINOv2 with ViT-g sets new accuracy records for self-supervised visual feature learning on multiple downstream tasks.
Noisy Student training with EfficientNet-L2 achieves state-of-the-art accuracy on ImageNet using self-training.
Baidu's ERNIE 3.0 Titan achieves high accuracy on Chinese and English benchmarks by incorporating knowledge graph embeddings.
RoBERTa-Large improves upon BERT with more training data and longer training, achieving higher accuracy on GLUE and other benchmarks.
Llama 3 70B is a powerful open-source large language model developed by Meta. It distinguishes itself through its massive training dataset and optimized architecture, resulting in exceptional performa...
Swin-L introduces shifted windows for efficient attention, achieving top accuracy on ImageNet and other vision tasks.
Vision Transformer Large achieves competitive accuracy on ImageNet by applying transformer architecture directly to image patches.
BERT-Large set new accuracy records on eleven NLP tasks, including question answering and language inference.
ConvNeXt-XL modernizes the standard ConvNet to achieve accuracy competitive with vision transformers on ImageNet.
