Welcome back to AI Weekly Updates. Each week we highlight the most important stories in AI so you can stay ahead without wading through endless headlines. From OpenAI shipping two frontier models in a week and a Florida criminal investigation, to Anthropic locking in $100 billion of AWS compute and Meta’s full-scale talent raid on Thinking Machines, here are the developments that shaped the AI landscape last week.
OpenAI ships ChatGPT Images 2.0 and a new gpt-image-2 API
OpenAI released ChatGPT Images 2.0 on April 21 — a new image generation model with improved text rendering, multilingual support, and visual reasoning that can pull live information from the web. The underlying gpt-image-2 model ships day-one in the API and Codex for developers. The upgrade targets the gap between a prompt and a finished editorial layout — posters, pitch decks, and documents that used to require a designer.
Read more ↗Anthropic commits $100B to AWS, secures up to 5GW of new compute
Anthropic and Amazon expanded their partnership on April 20 with a ten-year, $100 billion AWS commitment and up to 5 gigawatts of new training and inference capacity. Amazon is putting $5 billion in now with up to $20 billion more ahead, on top of $8 billion already invested. The framing is scale planning: Anthropic’s run-rate revenue hit $30 billion in early April 2026, up from roughly $9 billion at the end of 2025, and consumer demand has been straining Claude’s reliability at peak.
Read more ↗Anthropic launches Claude Design, taking on the Figma-and-Canva stack
Anthropic Labs shipped Claude Design on April 17 — a prompt-to-prototype product powered by Opus 4.7 that targets founders, PMs, and marketers who never opened a design tool. Canva integration ships day-one; a handoff bundle to Claude Code closes the loop from idea to production code. We wrote a deeper take on what it means for the design job market.
Read more ↗Google wires Gemini’s Nano Banana image model to your Google Photos
Google rolled out a Nano Banana update on April 16 that lets the Gemini app generate images using a user’s Google Photos library — no manual uploads, no long prompts. Personal context flows into the model automatically, enabling edits that reflect specific people, pets, and places from the user’s own life. The tie between Photos and the generator is Google’s answer to OpenAI’s editorial-image push: personal recall is the moat that OpenAI doesn’t have.
Read more ↗Florida AG opens criminal investigation into OpenAI over FSU shooting
Florida Attorney General James Uthmeier announced a criminal investigation into OpenAI on April 21, issuing subpoenas seeking records of how ChatGPT handles user threats of harm. The escalation from a previously civil probe stems from the 2025 Florida State University mass shooting, where Uthmeier claims ChatGPT “advised the shooter on what type of gun to use” and which ammunition paired with which weapon. It is the first time a state AG has treated an LLM’s outputs as potentially criminal.
Read more ↗Anthropic briefly pulls Claude Code from the $20 Pro tier, then reverses
On April 21, Anthropic removed access to Claude Code from its $20 Pro subscription on pricing pages and support documentation, before reversing course hours later after public pushback. Head of Growth Amol Avasare called the change “a small test of 2% of new prosumer signups,” but the support-doc edits that swapped “Pro or Max” to “Max” alone suggested a broader repricing signal. Current subscription plans “weren’t built” for recent usage patterns, Avasare said — expect restructuring across tiers.
Read more ↗OpenAI launches GPT-Rosalind, a frontier reasoning model for life sciences
OpenAI introduced GPT-Rosalind on April 16, a life-sciences reasoning model built for evidence synthesis, hypothesis generation, and experimental planning across biology, chemistry, and genomics. Launch partners include Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific. Named after Rosalind Franklin, the model ships as a research preview via a trusted-access program, with a free Life Sciences plugin for Codex that connects to more than 50 scientific tools and databases.
Read more ↗Meta hires five Thinking Machines founders after a $1B buyout was rejected
Meta has hired five of the founding members of Thinking Machines Lab, the AI startup Mira Murati built after leaving OpenAI, following Mark Zuckerberg’s rejected $1 billion acquisition offer. Co-founder Andrew Tulloch joined Meta Superintelligence Labs with a compensation package reportedly worth $1.5 billion over six years — if accurate, the most expensive individual talent hire in tech history. Of the startup’s founding group, five have now gone to Meta, three have returned to OpenAI, one joined xAI.
Read more ↗Meta pivots from open-weights Llama with closed-source Muse Spark
Meta Superintelligence Labs released Muse Spark on April 8 — its first AI model in a year and a closed alternative to the open-weights Llama strategy Meta had defined itself by since 2023. The natively multimodal reasoning model supports 262,000-token inputs, three reasoning modes, and tool use, landing fourth on Artificial Analysis’ Intelligence Index. Meta is withholding parameter count, architecture, and training details — a meaningful posture shift.
Read more ↗Meta plans to lay off 10% of its workforce — roughly 8,000 staff — in May
Meta is preparing to cut approximately 8,000 employees on May 20, per a Reuters report surfaced April 17, with additional cuts expected in the second half of the year. The move accompanies Meta Superintelligence Labs’ aggressive scaling under new chief AI officer Alexandr Wang, and analysts project up to $10 billion in annual savings redirected to AI infrastructure. The AI-first restructuring Zuckerberg telegraphed in 2023 is still compounding three years on.
Read more ↗