
Google released Gemma 4 12B, a 12-billion-parameter AI model that runs on laptops with 16GB of RAM. It matches the performance of larger Gemma 4 models in multistep reasoning and agentic workflows. The model uses Multi-Token Prediction drafters for efficiency, enabling faster inference without sacrificing capability. It occupies half the memory of the 26B MoE model and is available under Apache 2.0 license.
Tap to vote and see what everyone thinks.
Microsoft targets Anthropic with new model releases
Summary by ByteBrief