Announcing Gemma 3n, Imagen 4, and more! - Google Developer News June 2025

Key Concepts:

• Gemini 2.5 Flash-Lite & Pro: High-throughput and intelligent language models.
• Gemma 3n: Lightweight open model for mobile devices.
• Google Colab (AI-first): AI-powered coding environment.
• Imagen 4: Advanced text-to-image generation model.
• Grounding with Google Search: Enhancing model accuracy with real-time information.
• Code Execution: Model's ability to run and interpret code.
• Function Calling: Model's ability to use external functions or APIs.
• High-throughput tasks: Tasks that require processing a large volume of data quickly.
• Time to first token: The time it takes for a language model to generate the first word or token in a sequence.
• Parameter size: The number of parameters in a machine learning model, which affects its complexity and performance.
• Agentic collaborator: An AI assistant that actively collaborates with users on tasks.

Gemini 2.5 Model Family Updates

• Gemini 2.5 Flash-Lite: Designed for high-throughput tasks, emphasizing speed and efficiency. It offers a lower time to first token, leading to a more fluid user experience.
• Enhanced Capabilities: Gemini 2.5 Flash-Lite now supports grounding with Google Search, code execution, and function calling, improving its utility and accuracy.
• Availability: The stable release of Gemini 2.5 Flash-Lite is available for developers to integrate into their applications.
• Model Selection: Gemini 2.5 Pro remains the preferred choice for tasks requiring maximum intelligence, while Gemini 2.5 Flash is optimized for speed and cost-effectiveness.
• Pricing: A simpler and more predictable pricing structure has been implemented for the Gemini models.

Gemma 3n Open Model

• Lightweight Design: Gemma 3n is designed to run efficiently on mobile devices with limited memory (2-3 GB).
• Real-time Processing: It can process audio, video, and text in real time, making it suitable for interactive AI applications.
• Hardware Collaboration: Developed in collaboration with hardware manufacturers, incorporating architectural innovations for efficient performance.
• Adaptive Parameter Size: Gemma 3n can adjust its effective parameter size (4 billion) to optimize performance based on the task at hand.
• Availability: Open-model weights are available on platforms like Hugging Face, Kaggle, Ollama, and LM Studio.
• Developer Tools: Developers can use various tools to build and fine-tune models, including Google AI Studio and the Google AI Edge Gallery.

AI-First Google Colab

• AI Coding Expert: The upgraded Google Colab features an AI coding expert powered by Gemini models, providing real-time assistance to users.
• General Availability: The new AI-first Google Colab is now available to all users without a waitlist.
• Deep Integration: The AI assistant understands the user's code, actions, intentions, and goals, enabling more effective collaboration.
• Enhanced Productivity: Colab's agentic collaborator helps with code generation, error fixing, and data science tasks, accelerating the coding process.

Imagen 4 Text-to-Image Model

• Advanced Image Generation: Imagen 4, available in the Gemini API and Google AI Studio, generates sharper, more creative, and more controllable AI-generated visuals.
• Versatility: The model supports various styles and content, making it suitable for prototyping UIs, experimenting with concepts, and creative projects.
• Example: The model generated a three-panel sci-fi comic strip with words and a cohesive style from a simple prompt ("kapow").

Conclusion

The Google Developer News update highlights advancements in AI models and developer tools. Gemini 2.5 offers improved performance and capabilities, while Gemma 3n brings AI to mobile devices. The AI-first Google Colab enhances coding productivity, and Imagen 4 provides powerful text-to-image generation. These updates empower developers to build more innovative and efficient applications.