On Nov 11, Meta released MetaEncoder v1, a multimodal embedding model for image, video, and document retrieval.
Timeline · newest first
- Official
Meta releases MetaEncoder v1, a multimodal embedding model for image, video, and document retrieval
Hugging Face @facebook published this on Nov 11, 2026 (source: huggingface.co).
Meta published facebook/meta-encoder-v1 on Hugging Face at 2026-11-12T02:00:27Z (11 November 21:00 ET). The MuseGlimmer model embeds text, images, video frames, and documents, and exposes encode, retrieve, and decide calls that return 6,656-dimensional embeddings. It requires transformers 5.15 or later, the first release with the muse_glimmer architecture, and fits on one 80 GB GPU in bf16.
Source: Hugging Face @facebook
Related stories