What AI says about EVA‑CLIP‑18B

BAAI's EVA-CLIP-18B is an 18-billion-parameter open-source CLIP model for vision-language understanding, released on Hugging Face as the largest and most powerful open-source CLIP to date. It achieves 80.7% zero-shot top-1 accuracy across 27 image-classification benchmarks, trained on about 6 billion samples. The model is available for feature extraction via Hugging Face Transformers (with PyTorch support) and is part of the EVA-CLIP family alongside EVA-CLIP-8B.