SpaceX acquires Cursor for $60 billion → SpaceX has exercised its option to buy the popular AI coding tool Cursor in an all-stock deal. The two companies have already been jointly training frontier models together, with plans to release improved versions inside both Cursor and Grok Build. Read more here →

Midjourney launches portable full-body ultrasound → Midjourney is moving into healthcare with a new system that can generate full-body 3D ultrasound scans in about 60 seconds. The portable scanner needs no radiation or magnets and is built to work outside traditional hospital settings. Read more here →

Z.ai drops GLM-5.2 - SOTA open-weights model → Z.ai released GLM-5.2 under an MIT license, bringing big improvements in coding and long-context work up to 1 million tokens. The model is already topping open-source leaderboards for frontend coding and offers two different reasoning modes. Read more here →

Noam Shazeer, co-author of the landmark “Attention Is All You Need” paper that introduced the Transformer architecture, is joining OpenAI. → He previously co-founded Character.AI, contributed heavily to Google’s LaMDA and Gemini efforts, and is a pioneer in sparse Mixture-of-Experts models. OpenAI called the hire a major addition. Read more here →

Anthropic upgrades Claude Design with live editing → Claude Design now pulls in real design systems from codebases, checks its own output against them, and lets users edit directly on the canvas with drag-and-drop controls. Two-way syncing with Claude Code is also rolling out today. Read more here →

xAI ships Grok Imagine Video 1.5 → xAI launched Grok Imagine Video 1.5 with noticeably sharper motion, better physics, and much faster generation. 720p videos now finish in around 25 seconds and the model is available through both the API and consumer access. Read more here →

DeepSeek raises $7.4 billion at over $50B valuation → DeepSeek just closed a huge funding round at more than $50 billion, with its own CEO putting in $2.8 billion. The open-source company is using an unusual structure with long lockups while it keeps pushing its models forward. Read more here →

GPT-Realtime 2 shown as voice-controlled OS → A new demo shows GPT-Realtime 2 letting users control apps, browse the web, and even edit in Premiere Pro using just voice commands and simple setup. It connects through MCP and accessibility tools and feels like an early look at a voice-first way of using computers. Read more here →

Matt Pocock shares seven-phase AI workflow → Matt Pocock laid out a seven-phase approach to AI-assisted development that puts real emphasis on mapping the full design space early with tools like grill-with-docs before jumping into detailed work. The goal is to bring more structure to the messy pre-PRD stage. Read more here →