Gemma 4 Models: 31B and 26B MoE with 256K Context Window

Key Points:

• Google has released Gemma 4, an advanced open-source AI model family designed for diverse applications, featuring multi-modal integration of text, vision, and audio inputs, as well as enhanced long chain-of-thought reasoning for complex problem-solving.
• Gemma 4 offers two tiers: Workstation models optimized for high-performance tasks with a 256K context window, and Edge models tailored for resource-constrained devices with a 128K context window, supporting flexible deployment across industries.
• Released under the Apache 2.0 license, Gemma 4 encourages customization and collaboration, fostering innovation by providing developers full control over implementations in both enterprise and personal projects.
• The models demonstrate strong benchmark performance on industry standards like MMU Pro and SweetBench Pro, and support streamlined deployment via platforms such as Hugging Face and Google Cloud, facilitating integration into various infrastructures.
• With support for 140 pre-training languages and 35 fine-tuned languages, Gemma 4 is suited for multilingual and domain-specific applications, enabling advanced AI functionalities on everything from enterprise systems to edge devices like smartphones and IoT gadgets.