📄 Gemini can now create PDFs, Microsoft Word and Excel 📄
Gemini now exports Docs, Sheets, Slides, and PDFs without leaving the chat.

Google has rolled out native file-creation capabilities directly inside the Gemini chat interface. Users can now prompt Gemini to generate Google Docs, Sheets, Slides, Word documents, Excel spreadsheets, PDFs, CSV, LaTeX, and Markdown files — all downloadable instantly to their device. The update eliminates the copy-paste workflow that previously bridged Gemini outputs and productivity apps, a friction point that kept many users toggling between tools. By supporting both Google Workspace and Microsoft 365 formats, Google positions Gemini as the default creation layer across office ecosystems — deepening its distribution in enterprise workflows. The feature is rolling out now to all Gemini users.

🤖 Mistral Medium 3.5 powers cloud coding agents 🤖
Mistral's new 128B model runs asynchronous coding agents from the CLI or Le Chat.

French AI company Mistral has released Medium 3.5, a 128-billion-parameter dense model that powers a new "Vibe" remote-agent system for long, asynchronous coding tasks running in the cloud. Users can trigger tasks from the CLI or Le Chat's new "Work" mode, which chains multi-step actions across diverse tools. The model scores highly on SWE-Bench Verified while running on just four GPUs, balancing strong reasoning with practical inference costs. The release underscores Mistral's strategy of competing on agentic workflows rather than raw benchmarks — adding competitive pressure against US-based frontier labs. It also reinforces Europe's growing credibility in the foundation-model race.

💻 Claude Code adds push notifications for devs 💻
Anthropic tackles a key friction point in agentic coding rollout.

Anthropic has shipped push notifications for Claude Code, its developer-facing coding agent. The feature lets developers step away from the terminal and receive alerts when long-running tasks finish — a seemingly small UX update that addresses real friction in asynchronous coding workflows. It signals Anthropic's focus on making Claude Code practical for professional, multi-task environments, reinforcing the broader industry shift toward AI coding assistants as genuine background workers. The update reflects growing competitive pressure in the AI coding-agent space, where usability details increasingly drive adoption among engineering teams.

INTERESTING TO KNOW

🔬 DeepMind open-sources ProEval for cheaper evals 🔬

Google DeepMind has open-sourced ProEval, a framework designed to cut the cost of evaluating generative AI models. ProEval uses surrogate models and transfer learning to identify failure modes without full-scale evaluation runs — addressing a real pricing concern as eval costs increasingly rival training budgets. By making rigorous evaluation more accessible, the tool could level the playing field for smaller research groups and independent auditors. The framework is now available on GitHub.

📱 iOS 27 embeds generative AI into Apple Photos 📱

Apple is reportedly preparing three Apple Intelligence-powered editing tools for iOS 27, expected to be unveiled at WWDC on June 8. "Extend" will use generative AI to expand image backgrounds, "Enhance" auto-adjusts lighting and quality, and "Reframe" shifts perspective on spatial photos. The rollout marks Apple's deepest integration of on-device generative AI into its flagship Photos app, putting direct competitive pressure on Google's Magic Editor and other AI-powered mobile editing suites.