Google has rolled out Gemini Canvas — a persistent, Gemini‑powered workspace — to all users of AI Mode in Google Search in the United States in early March 2026. The feature, previously tested in Google Labs, appears as a side panel alongside Search’s chat and can be used on desktop and mobile (English only for now). Canvas lets users assemble and iterate on projects over time: planning trips with live flight and hotel data, drafting documents, building study guides, and generating working code or simple app prototypes that can be previewed, edited or exported. Google says Canvas pulls in real‑time Search data, Google Maps information and web sources, and supports creative writing, coding and lightweight tool creation. The rollout expands access to capabilities already available in Google’s Gemini app and to paid Gemini subscribers using larger model context windows. Google has not given a timeline for broader language or country expansion. Tech outlets and reviewers demonstrated rapid prototype creation, live data integration and on‑the‑fly code editing during the March announcements.

Google on April 16–17 rolled out a Personal Intelligence integration that lets its Gemini chatbot use the Nano Banana 2 image model to generate personalised images drawing on a user’s Google data, including Google Photos labels and other connected apps. Opt-in users who connect Photos, Gmail, Drive, Calendar, Maps and other account services can ask Gemini simple prompts — for example, “Design my dream house” or “Create a picture of my family” — and Gemini will automatically use inferred tastes, labelled photos and account context to shape outputs. The company says the feature will appear for paid AI subscribers (Plus, Pro and Ultra) in the United States in the coming days and will reach Chrome desktop and more users thereafter; Europe is not in the initial rollout. Google emphasises that Personal Intelligence is opt-in, that users can view a “sources” list showing which images informed a result, and that the Gemini app does not “directly” train models on private Photos libraries, though it may use prompts and responses to improve functionality. Users can refine results, provide feedback or select reference photos manually.

Kering CEO Luca de Meo told Reuters that Gucci will launch AI-powered smart glasses in partnership with Google in 2027, marking a major luxury-brand entry into the nascent smart-eyewear market. The product is expected to run on Google’s Android XR platform, joining other Google-backed collaborations such as Warby Parker and Gentle Monster. Details on design, features, pricing and exact launch timing remain limited; Google did not immediately comment. The move forms part of de Meo’s strategy to expand Kering’s eyewear and jewellery divisions and help revive Gucci’s sales, and it positions Gucci-Google devices directly against Meta’s Ray-Ban line and other anticipated offerings from Samsung, Apple and niche makers. The announcement was made during Kering’s capital markets day in Florence, Italy, and follows growing industry belief that stylish branding is key to mainstream adoption of smart glasses.

Anthropic on April 16-17 released Claude Opus 4.7, its most capable generally available model to date, positioning it as a production-ready AI for advanced software engineering, long-running agentic workflows and higher-resolution vision tasks. Anthropic says Opus 4.7 improves instruction following, autonomous multi-step reasoning, and memory for file-based workflows; it supports images up to 2,576 pixels on the long edge and retains a 1‑million token context. The company published benchmark gains on agentic coding and knowledge-work tests versus prior public models and kept API pricing unchanged at $5 per million input tokens and $25 per million output tokens. Opus 4.7 is available across Claude products and through partner clouds including Amazon Bedrock, Google Vertex AI and Microsoft Foundry. Anthropic also says it deliberately “differentially reduced” Opus 4.7’s cyber-offensive capabilities and added safeguards that block high‑risk cybersecurity requests, as it continues limited, partner-only testing of a more powerful frontier model, Claude Mythos Preview, under Project Glasswing.

Netflix will roll out a redesigned mobile app that includes a vertical video discovery feed by the end of April 2026, the company said in its Q1 2026 shareholder letter. The feature, tested in beta since early 2025, places short-form, scrollable clips and trailers on the app home screen so users can tap to watch, add titles to their list or share content. Netflix said the feed will be available across subscription plans; the company did not fully detail all interactive capabilities at launch. The update follows broader product moves including a standalone kids’ gaming app, Netflix Playground, with a worldwide rollout set for April 28. The vertical feed puts Netflix in line with rivals that have adopted swipeable short-video formats, such as Disney+ and Peacock, and reflects the company’s view that mobile consumption and video podcasts are blurring lines with TV viewing.

OpenAI on April 16-17, 2026 unveiled GPT‑Rosalind, a domain‑specific artificial intelligence model tailored for life sciences research and early‑stage drug discovery. Named for Rosalind Franklin, the model is designed to support evidence synthesis, hypothesis generation, experimental planning and multi‑step scientific workflows across biochemistry, genomics, protein engineering and translational medicine. OpenAI said GPT‑Rosalind is available as a research preview inside ChatGPT, Codex and the API but access is restricted to a vetted trusted‑access programme for qualified enterprise customers. The company is also releasing a free Life Sciences research plugin for Codex that links models to more than 50 scientific tools and databases. Launch partners and collaborators include Amgen, Moderna, Thermo Fisher Scientific, the Allen Institute, Los Alamos National Laboratory and Dyno Therapeutics. OpenAI published benchmark results showing a 0.751 pass rate on BixBench and outperformance of GPT‑5.4 on several LABBench2 tasks; a partner evaluation on unpublished RNA sequences reportedly placed model outputs above the 95th percentile of human experts on certain prediction tasks. OpenAI said technical and governance safeguards will be used to limit misuse and that the preview is aimed at organisations working to improve human health outcomes.