Google Releases Gemma 4 Under Apache 2.0 — Open-Source AI Gets a Commercial-Grade Upgrade

Google DeepMind released Gemma 4 on April 2, 2026, and the combination of performance and licensing makes it one of the most significant open model releases of the year. Built from the same research underlying Gemini 3, Gemma 4 ships in four sizes — from a 2B edge model to a 31B dense model — with context windows up to 256K tokens, native vision and audio processing, and fluency in over 140 languages. The 31B model jumped from 20.8% to 89.2% on the AIME math benchmark and from 29.1% to 80.0% on LiveCodeBench compared to Gemma 3.

The Apache 2.0 License Changes Everything

Previous Gemma releases shipped under Google's custom license, which imposed restrictions on commercial use and redistribution. Gemma 4 moves to the fully permissive Apache 2.0 license, enabling unrestricted commercial use, modification, and redistribution with no royalty obligations. For enterprises building AI-powered products, this removes the legal uncertainty that kept some teams from adopting open models. Companies can now fine-tune Gemma 4 on proprietary data and deploy it in production without navigating complex licensing terms.

Edge AI Gets Practical

The smaller Gemma 4 variants are designed to run entirely on-device — phones, Raspberry Pi units, and NVIDIA Jetson boards — with near-zero latency and no internet connection required. The 2B and 4B effective parameter models use mixture-of-experts architecture to activate only a fraction of their parameters during inference, preserving RAM and battery life. For businesses in healthcare, manufacturing, and field services where cloud connectivity is unreliable or prohibited, this opens new deployment scenarios that were previously impractical.

What This Means for AI Professionals

Gemma 4 narrows the gap between what companies can build with open models versus proprietary APIs. Teams with experience fine-tuning and deploying open models are increasingly valuable as enterprises weigh the total cost of ownership between API-based and self-hosted AI. Professionals who understand model quantization, edge deployment, and on-premise inference infrastructure are positioned for growing demand as more organizations bring AI workloads in-house.