GOOGLE GEMMA 4: THE OPEN-WEIGHTS GAME CHANGER

Google DeepMind’s Gemma 4, released April 2, represents a significant shift in the open-weights AI landscape. The model suite — available in four sizes — is built from the same technology as Gemini 3 but released under Apache 2.0 license. This means completely unrestricted commercial use, a notable departure from more restrictive open-source licenses.

Key capabilities:

• Competitive with frontier models on benchmarks
• Apache 2.0 license (truly open for commercial deployment)
• Four size variants for different use cases
• Built from Gemini 3 technology

The timing is strategic: Microsoft announced three in-house models on the same day, and Anthropic’s Claude is facing criticism for blocking third-party agents. Developers looking for alternatives now have a compelling open option.

For the AI ecosystem, Gemma 4 changes the calculus:

• Startups can deploy powerful AI without API costs
• Enterprises can run models on-premise with full control
• Researchers can study and modify model weights freely
• Competition intensifies for paid API providers

Google’s approach contrasts with Anthropic’s tightening: while Claude is becoming more restrictive, Gemma 4 offers more freedom. The open-weights movement is gaining momentum at exactly the moment when frontier providers are tightening control.

Sources:

• Google DeepMind: Gemma 4 announcement
• Gigazine: Apache 2.0 license details
• Technology analysis reports