Gemma 3 was released more than a year ago, in March of 2025. Following this, Google has introduced Gemma 4, a model designed for advanced reasoning, coding, and multimodal workflows. This new model is available in four sizes: Effective 2B (E2B), Effective 4B (E4B), a 26B Mixture of Experts model, and a 31B dense model.

Gemma 4 supports multimodal processing, enabling it to work with text, images, and video inputs. This capability allows the AI to generate code, process images and videos, and understand speech across more than 140 languages. The model is built using the same research behind Google’s Gemini series, aiming to achieve multi-step reasoning and manage highly complex tasks.

Released under the Apache 2.0 license, Gemma 4 is accessible for both commercial and research use. The Gemma ecosystem has seen significant adoption, surpassing 400 million downloads since its initial launch. This widespread use underscores the demand for advanced AI solutions in various sectors.

Initial reactions to Gemma 4 highlight its efficiency, with statements noting that the core idea is to deliver strong AI performance without requiring high-performance computing. This efficiency enables these models to compete with much larger AI systems while utilizing significantly fewer resources.

Details remain unconfirmed regarding the performance of Gemma 4 in comparison to other models. Additionally, there are rumors about a larger Mixture of Experts model with over 100B total parameters that has not yet been released.