Google DeepMind Releases Gemma 4 12B: Unified Encoder-Free Multimodal Model

Google DeepMind released Gemma 4 12B, an open-weight multimodal model notable for its encoder-free architecture that handles text and vision in a single unified model without a separate vision encoder. The 12B parameter size targets on-device and local deployment use cases where frontier-scale models are impractical. As an open-weight release, Gemma 4 12B is available for fine-tuning and local deployment, continuing Google’s effort to compete with Meta Llama and Mistral in the open-weight segment.