Google Releases Gemma 4: Most Capable Open Models with Apache 2.0 License

TLDR

Google has released Gemma 4, described as 'our most intelligent open models to date, purpose-built for advanced reasoning and agentic workflows.' The lineup includes models ranging from 2 billion to 31 billion parameters, supporting over 140 languages with multimodal capabilities for processing text, video, and audio inputs.

A major change involves switching from Google's restrictive license to Apache 2.0, removing previous limitations on model deployment and addressing enterprise concerns about vendor lock-in. The flagship 31B model can run on a single 80GB H100 GPU at full precision or on consumer hardware like an RTX 4090 with 4-bit quantization, making frontier-level capabilities accessible to the broader development community.

Key Takeaways

• Google DeepMind released Gemma 4, a family of multimodal open models from 2B to 31B parameters supporting 140+ languages, now under Apache 2