code.google.com

Gemini 3 Flash is now available in Gemini CLI. It delivers Pro-grade coding performance with low latency and a lower cost, matching Gemini 3 Pro's SWE-bench Verified score of 76%. It significantly outperforms 2.5 Pro, improving auto-routing and agentic coding. It's ideal for high-frequency development tasks, handling complex code generation, large context windows (like processing 1,000 comment pull requests), and generating load-testing scripts quickly and reliably.

LiteRT's new Qualcomm AI Engine Direct (QNN) Accelerator unlocks dedicated NPU power for on-device GenAI on Android. It offers a unified mobile deployment workflow, SOTA performance (up to 100x speedup over CPU), and full model delegation. This enables smooth, real-time AI experiences, with FastVLM-0.5B achieving over 11,000 tokens/sec prefill on Snapdragon 8 Elite Gen 5 NPU.

Google Colab has launched an official VS Code extension, bridging the gap between the popular code editor and the AI/ML platform. The extension combines VS Code's powerful development environment with Colab's seamless access to high-powered runtimes (GPUs/TPUs), allowing users to connect local notebooks to Colab. This aims to meet developers where they are and brings the best of both worlds.

Code Wiki is a new platform that tackles the bottleneck of reading existing code by providing an automated, continuously updated, structured wiki for code repositories. It features hyper-linked documentation, a Gemini-powered chat agent that understands your repo, and automated diagrams. A public preview is available for open-source projects, and a Gemini CLI extension is coming soon for secure use on private repos.

Gemini 3 Pro is now integrated into Gemini CLI, unlocking state-of-the-art reasoning, agentic coding, and advanced tool use for enhanced developer productivity. It's available now for Google AI Ultra and paid Gemini API key subscribers (upgrade CLI to 0.16.x). Features include generating 3D apps and code from visual sketches, running complex shell commands, creating documentation, and debugging live Cloud Run services.

Jules, an always-on, multi-step software development agent, now features Gemini 3, offering clearer reasoning and better reliability. Recent improvements include parallel CLI runs, a stable API, and safer Git handling. Upcoming features include directory attachment without GitHub and automatic PR creation. Jules aims to reduce software writing overhead so developers can focus on building.

The JAX AI Stack is a modular, industrial-grade, end-to-end machine learning platform built on the core JAX library, co-designed with Cloud TPUs. It features key components like JAX, Flax, Optax, and Orbax for foundational model development, plus an extended ecosystem for the full ML lifecycle and production. This integration provides a powerful, scalable foundation for AI development, delivering significant performance advantages.

Gemini 3 Pro Preview is introduced as a powerful, agentic model for complex, (semi)-autonomous workflows. New agentic features include `thinking_level` for reasoning control, Stateful Tool Use via Thought Signatures, and `media_resolution` for multimodal fidelity. It has Day 0 support for open-source frameworks like LangChain, AI SDK, LlamaIndex, Pydantic AI, and n8n. Best practices include simplifying prompts and keeping temperature at 1.0.

Introducing Google Antigravity, a new agentic development platform for orchestrating code. It combines an AI-powered Editor View with a Manager Surface to deploy agents that autonomously plan, execute, and verify complex tasks across your editor, terminal, and browser. Agents communicate progress via Artifacts (screenshots, recordings) for easy verification. Available now in public preview.

Gemini 3 is available via API with updates for developers: new `thinking_level` for depth control, `media_resolution` for multimodal processing, and enforced `Thought Signatures` for agentic workflows, especially with function calling and image generation. It also introduces combining Google Search/URL Grounding with Structured Outputs and new usage-based pricing for Grounding. Best practices, like using default temperature, are advised for optimal results.