1 of 1
Story summary
- Google's Gemma 4 models balance quality and speed in local LLMs.
- Gemma 4 offers four distinct models based on the Gemini 3 research, all under Apache 2.0.
- The lineup includes a 31B dense model for high quality and a 26B-A4B Mixture of Experts model that activates only 3.8 billion parameters per token for faster performance.
- Each model is built differently but shares a training approach to enhance usability.
