Model5h ago
Google DeepMind Releases Gemma 4 12B
The 12B parameter Gemma 4 utilizes an encoder-free architecture to process text, images, and audio natively. This unified approach removes separate modality encoders, reducing latency and memory overhead. Google DeepMind optimized the model for high-performance edge deployment. Developers can now run complex multimodal reasoning on consumer-grade hardware without sacrificing accuracy or speed.