Google has announced Gemma 4, its latest family of open AI models, and the headline feature isn’t a benchmark improvement or a new capability — it’s the license. Gemma 4 moves from Google’s custom restrictive license to the Apache 2.0 license, one of the most permissive open source licenses in existence.
Why the License Change Matters
Previous versions of Google’s Gemma models used a custom license that drew significant criticism from the open source community. The restrictions embedded in that license effectively limited what developers could do with the models — particularly for commercial applications — without requesting special permission from Google.
The Apache 2.0 license changes everything. It is the same license used by Google for Android and countless other open source projects. Under Apache 2.0:
- Anyone can use the models for any purpose, including commercial applications
- Developers can modify, distribute, and create derivative works
- No royalties or special permissions are required
- The license provides patent protection from Google
Performance Improvements
Beyond the licensing change, Gemma 4 brings performance improvements over its predecessors. The Gemma family has consistently been Google’s answer to Meta’s Llama series — compact, capable open weight models that developers can run locally or fine-tune for specific use cases.
Gemma 4 continues this tradition with better reasoning capabilities, improved multilingual performance, and more efficient inference. For developers who have been waiting for a Google model they could truly use without legal ambiguity, this release is significant.
Open Source AI: The Licensing Wars
Gemma 4’s move to Apache 2.0 reflects a broader trend in the AI industry: the battle over what “open source AI” really means. Meta’s Llama models have been praised for their permissive licensing but criticized for not being truly open — the weights are released, but the training data and infrastructure remain proprietary.
Google’s Gemma 4 takes a different approach. By switching to Apache 2.0, Google is making a clear statement: this is what we mean when we say open. No usage restrictions. No commercial limitations. No royalty negotiations. Just open weights, Apache 2.0, and the full backing of Google’s intellectual property portfolio.
What This Means for Developers
For the developer community, Gemma 4 removes a major pain point. Enterprises that were hesitant to build on Gemma 3 due to licensing uncertainty can now move forward confidently. Research institutions can fine-tune Gemma 4 for specific domains without legal review. Startups can embed Gemma 4 in commercial products without budget allocations for licensing fees or negotiations with Google.
The combination of genuinely open licensing with Google’s model architecture creates a new baseline for what open source AI can be. As more companies jockey for position in the open source AI space, Google’s move with Gemma 4 may force competitors to reconsider their own licensing strategies.
The Bigger Picture
Google’s decision to open up Gemma 4 is also a strategic move in the AI race. By making Gemma 4 genuinely open, Google positions itself as a friend to the developer community — a narrative that has been complicated by Google’s history of proprietary products and services.