DeepMind introduces Gemma 4 12B, a unified, encoder-free multimodal model
HIGHLIGHTS
- •DeepMind launched Gemma 4 12B, a new multimodal model
- •The model features a novel unified architecture without multimodal encoders, allowing direct integration of audio and vision inputs into the LLM backbone. It is designed to run on consumer laptops with 16GB of RAM and is released under an Apache 2.0 license.
- •Gemma 4 12B brings high-performance multimodal intelligence to everyday hardware without sacrificing speed or reasoning, enabling advanced multimodal and agentic experiences directly on consumer devices.
Published:
Official Source (Official Blog)