NeoMME: an efficient Multimodal-native and Multilingual Encoder

What changed

We introduce NeoMME, a family of 260M and 800M multilingual multimodal encoders. Unlike many generative visual language models, NeoMME does not use a separate pretrained vision tower or a causal language model. A single bidirectional Transformer processes both text tokens and raw image patches, and we train the entire model from scratch with a masked discrete-diffusion objective.

Why it matters

A concrete addition to Practical AI: Tools, Models & Frameworks: it changes what's available to builders today rather than being general commentary.

How it compares

Related prior coverage to compare against:

• Efficient MultiModal Data Pipeline
• Introducing Gemma 4 12B: a unified, encoder-free multimodal model
• Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM

• Efficient MultiModal Data Pipeline similarity 0.32
• Introducing Gemma 4 12B: a unified, encoder-free multimodal model similarity 0.30
• Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM similarity 0.26