All Models

CLIP ViT-L/14 at 336-pixel resolution (April 2022), the most accurate model in OpenAI's CLIP release, for zero-shot image classification and image-text matching; slower than ViT-B/32.

OpenAI's original CLIP ViT-B/32 (January 2021): a contrastive image-text model that classifies images zero-shot against any text labels; small and fast, but superseded for most uses by newer encoders such as SigLIP 2.