LightOn ships NeoMME: natively multimodal encoders at 260M/800M with no vision tower

antoine_chaffin · x · 2026-09-03

LightOn releases NeoMME, rejecting the common practice of using massive generative VLMs as representation models. Instead, the team trained encoders from scratch that are natively text-and-image, multilingual, and built for speed from the start — at just 260M and 800M parameters, with no vision tower.

Related event: LightOn Unveils NeoMME: Single-Tower Multimodal Encoder That Leads Sub-800M Retrieval(4 posts)→

Original post →

More from Multimodal

Multimodal channel →