Ettin Rerankers offered more size choices for a second retrieval pass
Six cross-encoder models arrived with data and a training recipe for examining the quality-efficiency tradeoff.
Article·4 minEXPLORE THE ECOSYSTEM
Models
Small language models, multimodal releases, and the open-weight ecosystem.
18 articlesNewest announcements first
Article·4 min
Sentence Transformers extended retrieval beyond text
The library added a shared interface for comparing text, images, audio, and video.
Article·4 min
Gemma 4 brought another wave of multimodal models to the Hub
The family’s release connected multimodal capabilities with a broad set of open-model tools.
Article·4 min
Waypoint-1 explored video generation as an interactive world
Overworld’s model used user controls as part of the generation process instead of only producing a fixed clip.
Article·4 min
FLUX.2 arrived as a new pipeline, not a simple FLUX.1 swap
The release combined image generation and editing with a different architecture and multiple-reference inputs.
Article·3 min
EmbeddingGemma focused on compact multilingual retrieval
A small embedding model targeted local search and retrieval-oriented applications.
Article·4 min
SmolLM3 paired a small model with a detailed training recipe
The three-billion-parameter release emphasized multilingual capability, reasoning modes, and transparency.
Article·4 min
Gemma 3n brings multimodal models closer to local devices
The full release connects an on-device model family to familiar open tooling.
Article·4 min
Gemma 3 offered a wider range of multimodal deployment choices
The family expanded across model sizes, languages, and context lengths, with important differences between variants.
Article·4 min
SigLIP 2 improved the building blocks of multimodal search
Google’s vision-language encoders combined multilingual training with stronger visual representations.
Article·4 min