Google's Gemma 4 Goes Open Source with Apache 2.0, Rivals Top AI Models

Google Opens the Floodgates with Gemma 4

In an early morning announcement on April 3, Google DeepMind unveiled its latest open-source AI model series - Gemma 4. What makes this release particularly noteworthy isn't just the technical specs (though those are impressive), but Google's decision to switch from proprietary licensing to the widely-adopted Apache 2.0. This move removes commercial use restrictions, giving developers unprecedented freedom to build upon Google's work.

A Model for Every Need

Gemma 4 arrives not as a single model, but as a carefully curated family designed for different applications:

• 31B Dense (Flagship Edition): The heavyweight champion with full parameter activation and support for ultra-long context (256K). Surprisingly efficient, it runs on just one H100 GPU.
• 26B A4B MoE: The cost-efficient option using mixture-of-experts architecture - think of it as activating only the relevant parts of the model when needed.
• E-Series (Edge Elite): These compact models bring AI capabilities to mobile devices, with memory footprints as low as 1.5GB.

Performance That Turns Heads

The numbers tell a compelling story of progress:

• Math skills jumped from scoring 20% on advanced tests to nearly 90%
• Programming ability now rivals many human developers, with Codeforces ratings soaring past 2000
• Graduate-level science comprehension nearly doubled to 84.3%
• Native support for 140+ languages makes this truly global AI

The improvements aren't just quantitative - qualitative changes like built-in "thinking mode" allow the model to reason through problems before answering, much like its flagship Gemini counterpart.

More Than Just Parameters

What sets Gemma 4 apart goes beyond raw computational power:

• Agent-ready architecture: Built-in support for function calls and structured outputs means these models can power intelligent assistants right out of the box.
• Multimodal from the ground up: Even the smallest versions handle images, video, and audio - no afterthoughts here.
• Developer-friendly tools: Google released an accompanying Agent Development Kit (ADK) to help creators build practical applications.

Shaking Up Open Source AI

The timing of this release is strategic. As regional open-source models like China's DeepSeek gained momentum, Google needed to reaffirm its position in this space. By combining cutting-edge engineering with genuine openness (that Apache license matters), Gemma 4 represents both a technical achievement and a philosophical statement about accessible AI.

Key Points:

• Open access: Apache 2.0 license enables commercial use and modification
• Performance leap: Dramatic improvements in math, coding, and reasoning tasks
• Scalable options: From data centers to smartphones in one model family
• Future-ready: Native agent support and multimodal capabilities built in