Google Releases Gemma 4 Open-Source AI Models
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google released Gemma 4, a family of open-source AI models under the Apache 2.0 license, available on Hugging Face and Google Cloud for modification, fine-tuning, and commercial deployment
- Gemma 4 ships in two tiers: Workstation models (31B dense and 26B mixture-of-experts) with 256K context windows for demanding tasks, and Edge models (E2B and E4B) with 128K context windows for resource-constrained devices like smartphones, IoT devices, and Raspberry Pis
- Multimodal capabilities natively process text, vision, and audio inputs, with an enhanced vision encoder supporting aspect ratio processing and multi-image analysis
- Audio functionality includes transcription, translation, and speech recognition tuned for high accuracy in edge environments
- Gemma 4 demonstrated strong results on MMU Pro and SweetBench Pro benchmarks, leveraging long chain-of-thought reasoning for complex multi-turn agentic flows and function calling
- Deployment supports serverless scaling via Cloud Run with G4 GPUs, and the models support 140 pre-training languages for multilingual applications
Why it matters: The Apache 2.0 license lets developers and enterprises modify, fine-tune, and commercially deploy Gemma 4 without restrictive fees, lowering the barrier for organizations of all sizes to build on advanced AI. The dual-tier design — 256K context for cloud-heavy workflows and 128K context for edge devices — means the same model family can power both enterprise applications and lightweight on-device use cases.
