Google Launches Gemma 4 AI Models with Apache 2.0 License

Google has released Gemma 4, its latest open AI models, marking a significant shift in accessibility and licensing.

The new models come in four sizes tailored for local use. The largest variants, 26B Mixture of Experts and 31B Dense, are built for high-performance hardware like Nvidia H100 GPUs. However, they can also run on consumer GPUs when quantized.

Gemma 4 also introduces smaller models, Effective 2B (E2B) and Effective 4B (E4B), optimized for mobile devices such as smartphones and Raspberry Pi.

Key improvements include reduced latency and efficient parameter activation, especially in the 26B model, which uses only 3.8 billion of its 26 billion parameters during inference.

Google now offers the Gemma 4 models under the Apache 2.0 license, addressing developer frustrations with previous proprietary licensing.

This move positions Gemma 4 among the top open-source AI models, offering greater flexibility without compromising performance.