GOOGLE LAUNCHES GEMMA 4 12B MULTIMODAL MODEL

Gemma 4 12B eliminates the traditional encoder-decoder structure common in multimodal models, instead using a single unified architecture. This design choice reduces complexity while maintaining performance across text and image understanding tasks. The model operates at 12 billion parameters, positioning it as a mid-size option between smaller and larger language models. Google targets developers building applications requiring multimodal capabilities without the computational overhead of larger alternatives. Key features include direct image and text processing in a single forward pass and encoder-free architecture reducing model complexity. The release aligns with Google's strategy of providing modular AI tools for developers. Gemma 4 12B is available through Google's developer tools and platforms. The announcement generated 218 points and 85 comments on Hacker News, indicating developer interest in the efficiency-focused approach.