Google code homepage
Gemma 3 QAT Models: Bringing state-of-the-Art AI to consumer GPUs
The release of int4 quantized versions of Gemma 3 models, optimized with Quantization Aware Training (QAT) brings significantly reduced memory requirements, allowing users to run powerful models like Gemma 3 27B on consumer-grade GPUs such as the NVIDIA RTX 3090.
The Gemini API and the Internet of Things
The Gemini API and ESP32 microcontroller simplify custom voice commands for IoT devices, leveraging speech recognition for devices to understand and react to custom commands, bridging the gap between digital and physical worlds.
Leveraging BigQuery JSON for Optimized MongoDB Dataflow Pipelines
An enhancement to Google Cloud Dataflow templates for MongoDB Atlas enables direct integration of JSON data into BigQuery, eliminating complex data transformations, reducing operational costs, and enhancing query performance for users.
Safer and Multimodal: Responsible AI with Gemma
ShieldGemma 2, built on Gemma 3, is a 4 billion parameter model that can be used as an input filter for vision language models or an output filter for image generation systems, and is designed to respond to a wide range of diverse and nuanced imagery.
Experiment with Gemini 2.0 Flash native image generation
The experimental native image generation feature of Gemini 2.0 Flash – allowing for the combination of text and images, conversational image editing, and leveraging real-world knowledge for contextual visuals – is now available for developers to test through Google AI Studio and the Gemini API.
Introducing Gemma 3: The Developer Guide
Gemma 3 is a new, advanced version of the Gemma open-model family featuring multimodality, longer context windows, and improved language capabilities, with various sizes and deployment options for developers to experiment.
Gemma 3 on mobile and web with Google AI Edge
Gemma 3 1B, a new small language model for mobile and web applications via Google AI Edge, is now available, with increased efficiency, improved performance, and offline availability.
Unlocking bonus worlds with Gemini for the Google I/O puzzle
The Google I/O 2025 puzzle used the Gemini API to generate dynamic riddles for bonus worlds, enhancing player engagement and scalability. Here's what our developers learned on using the Gemini API effectively, including creativity, design, and implementation strategies.
Introducing TxGemma: Open models to improve therapeutics development
Google DeepMind releases TxGemma, built on Gemma, which predicts therapeutic properties, and Agentic-Tx, powered by Gemini 2.0 Pro, which tackles complex research problem-solving with advanced tools.
Simplified Dataflow Connectors with Managed I/O
Google Cloud Dataflow's Managed I/O simplifies using Apache Beam I/O connectors by automatically updating connectors to the latest versions and providing a standardized API, optimizing connectors specifically for Dataflow, ensuring efficient performance and reducing the need for manual configuration, freeing users to focus on pipeline logic.