Google DeepMind Launches Gemma 4 Under Apache 2.0 License, Removing Key Enterprise Barrier

Google DeepMind released Gemma 4 on April 2, 2026 under the Apache 2.0 license — a first for the Gemma model family. Prior Gemma releases were governed by Google's own usage policy, which restricted commercial redistribution and modification in ways that created friction for enterprise deployment. Apache 2.0 allows unrestricted commercial use, redistribution, and derivative works, removing the primary legal barrier that had prevented many enterprises from integrating Gemma into production systems.

The Gemma 4 release includes a family of models ranging from 2.3B to 31B parameters, with the flagship being a 26B mixture-of-experts variant that keeps only 4 billion parameters active per inference pass — delivering competitive performance at a fraction of the compute cost of equivalent dense models. Google also released a Gemma 4 multimodal variant capable of processing images alongside text, trained on Google's internal multimodal instruction datasets. All models are available on Google Cloud, Hugging Face, and Kaggle.

The license change is being widely described as the most strategically significant aspect of the release. Industry observers at 9to5Google and the Google Open Source Blog noted that Apache 2.0 positions Gemma 4 to compete directly with Meta's Llama series — which released the same week under its own community license — for enterprise adoption. Google Cloud simultaneously announced expanded fine-tuning infrastructure for Gemma 4, including one-click deployment on Vertex AI, as the company accelerates its push to make its open-weight models the preferred foundation for enterprise AI workloads.