Google releases its Gemini 3 model and unveils Antigravity, an agent-based coding platform that can autonomously execute tasks on a user’s computer. The launch moves the conversation beyond text generation to AI that plans, codes, and coordinates work with human oversight. In real-world tests, Gemini 3 built a playable game from a single prompt and created a full website that summarized years of blog posts, all while routing approvals through an inbox interface. Antigravity reads local files, writes code, conducts web research, and even controls the browser to validate its output. The model also cleaned messy research data, devised fresh hypotheses, executed statistical analysis, and delivered a 14-page journal-style paper with minimal guidance. The author says managing Gemini 3 feels like supervising a capable graduate assistant rather than coaxing a chatbot.
Google launched a redesigned Build mode in AI Studio that lets anyone generate and deploy a web app from a simple text prompt. The update, branded as “vibe coding,” is available now at ai.studio/build and requires no payment info to begin. Users can mix Gemini 2.5 Pro with tools like Veo, Imagine and Flashlight, edit the full React/TypeScript source, and push directly to GitHub or Cloud Run. An “I’m Feeling Lucky” button auto-creates app concepts for inspiration, while advanced models and Cloud Run deployment unlock only after adding a paid API key. The hands-on demo showed a novice building a working dice-rolling app in 65 seconds, highlighting how far the barrier to AI app creation has fallen. That speed and simplicity position Google’s offering as a direct challenger to developer-oriented tools like OpenAI’s Codex and Anthropic’s Claude Code, according to the article.
Kitten TTS is an open-source series of tiny and expressive text-to-speech models for on-device applications. We are excited to launch a preview of our smallest model, which is less than 25 MB. This model has 15M parameters.
This release supports English text-to-speech applications in eight voices: four male and four female. The model is quantized to int8 + fp16, and it uses onnx for runtime. The model is designed to run literally anywhere eg. raspberry pi, low-end smartphones, wearables, browsers etc. No GPU required!
The Canva Design Model understands structure and hierarchy to produce completely editable designs, with integration into ChatGPT, Claude, and Gemini. The Creative Operating System’s tools include Video 2.0 for streamlined editing, forms, data connectors, email design, and a 3D generator. Grow consolidates marketing workflows by letting teams browse winning ads, create brand-aware variations, publish directly to Meta, and track performance. Canva’s 2024 acquisition of pro-design tool Affinity is also relaunching as an all-in-one free creative app with built-in Canva integrations. Why it matters: AI design tools have come a long way in the past year, and Canva is keeping itself on pace with the acceleration. Now with its own model and an AI feature for every creative need, the disruptive platform is not only empowering its users, but also reducing the need to ever hop to other rivals or more ‘professional’ options.