Welcome back! OpenAI showed up Thursday to launch its new models and dropped a whole AI agent too, one that looks an awful lot like Claude. Meta put out a coding model at a quarter of everyone else's price. Ben Bernanke, who ran the Fed through the 2008 crash, joined the body that oversees Anthropic. And Google built a locked room where AI can run its own code without touching yours.
In today's Generative AI Newsletter:
OpenAI: What did OpenAI drop alongside GPT-5.6?
Meta: How cheap is Meta's new coding model?
Anthropic: Why did a former Fed chair just join Anthropic?
Google: Where is all that AI-written code supposed to run?
Thursday was supposed to be about the GPT-5.6 models going public. OpenAI used the moment to drop something bigger.
ChatGPT Work is an agent you hand a goal, and it works across your apps and files for hours before handing back a finished deck, sheet or dashboard.
On desktop it reads your local files, drives its own built-in browser, and clicks and types across your apps on its own.
It plugs into Slack, Gmail, Salesforce, GitHub and a dozen more, so you can point it at real work. One company turned it loose on their sales pipeline and it dug up seven figures in missed deals.
To cram all this in, OpenAI folded Codex into a new desktop app and renamed the old one ChatGPT Classic.
It's Claude Cowork with more muscle, and Anthropic saw it coming, rushing Cowork onto phones two days earlier and resetting all user limits yesterday. OpenAI's version digs deeper into your computer. It just wrecked the chat box getting there and stopped supporting their Atlas browser.
Special highlight from our network
Most professionals open Claude, ask a question, and close the tab. That's about 5% of what Claude can actually do in 2026.
The other 95% is where the interesting work is happening. Skills that give Claude persistent expertise. Connectors that plug it into Notion, files, and your desktop. Cowork for full document and project workflows. Vibe coding for building apps without being a developer.
Outskill's 2-Day Claude AI Mastery Workshop is a live end-to-end walkthrough of all of it, condensed from 800+ hours of research into a focused 16-hour curriculum. Plus 10+ AI tools and workflows that pair with Claude.
Saturday and Sunday, 10 AM to 7 PM EST. Free for the next 48 hours.
Building an AI product is easy now. Getting it in front of buyers is the hard part.
Companies like Nvidia, Oracle and Notion come to us to show their products to millions. We productized that knowledge to create a market-leading GTM app called ToneUp.
Revenue grew 2.5x last year. You can invest at an early shareholder price starting from $1,000 with up to 22% bonus shares.
Meta launched Muse Spark 1.1, a model built for agent work, coding and computer use, with a million-token memory and the ability to run a swarm of subagents at once.
It's fast, and it's cheap. $1.25 per million tokens in and $4.25 out, roughly a quarter of what OpenAI and Anthropic charge for comparable work.
Mark Zuckerberg announced it himself, his first post on X in three years.
It's the first model Meta has ever charged for.
The company gave Llama away free for years, ended that in April, and now bills by the token, and developers are already asking for the open weights it hasn't shared yet.
Replit's CEO called it "a complete agentic foundation." Grok undercut Claude on Wednesday, Meta undercut the field on Thursday, and Google keeps getting lapped in a race it used to lead.
Anthropic appointed Ben Bernanke to its Long-Term Benefit Trust, the independent body that oversees the company.
This is the man who ran the Federal Reserve through the 2008 crash, won the 2022 Nobel in economics for his work on how banks fail, and chaired Princeton's economics department. He holds no stock in Anthropic.
The trust has teeth. Its members can appoint and remove a majority of Anthropic's board, the exact power OpenAI's board turned out not to have when it tried to fire Sam Altman in 2023 and caved within days.
Anthropic filed to go public on June 1. Putting a former Fed chair on the body that can overrule your CEO is the kind of thing that makes Wall Street comfortable.
Bernanke said it best himself. How AI's potential plays out "will depend, in part, on the institutions we build around it." He just became one of them.
Every agent that writes code needs somewhere to run it without wrecking your machine. Google's answer is Cloud Run sandboxes, out in preview, a locked room that spins up in milliseconds and disappears when the job's done.
Security is the whole pitch. A sandbox has zero network access by default, so if an agent gets tricked into running a script that tries to phone home with your data, the request dies at the door. It can't reach your credentials either.
Google ran 1,000 of these in a demo at half a second each, and they cost nothing on top of what you already pay for Cloud Run.
This is the boring plumbing under every flashy launch this week. ChatGPT Work and Muse Spark both write and run code. This is where that code is meant to live so it can't hurt you.
Julius is a chat-based data analyst. You hand it a spreadsheet, ask a question in plain English, and it runs the analysis, builds the charts and puts together the slides. Good timing for a Friday when someone needs the week's numbers made sense of.
Try this yourself:
Sign up free and connect Google Drive or drag in a CSV or Excel file.
Ask it something real like "build me a model of this month's spend" and watch it work.
Turn the result into a slide deck or a chart you can paste straight into a report.
Free gives you 2,000 credits a month; Plus is $16 a month when you outgrow it.
Who it's for: anyone who lives in spreadsheets and would rather ask than write another formula.
OpenAI's No. 2 is stepping down: Fidji Simo, hired from Instacart to turn OpenAI's models into products, is leaving her full-time role for health reasons and staying on as an advisor, right as the company lines up an IPO.
Google's algorithm-inventing agent went public: AlphaEvolve, the DeepMind system that writes and improves its own code, hit general availability, and Klarna used it to double its machine-learning training throughput.
OpenAI and Google sold AI to blacklisted firms' offshoots: The FT found both supplied models to Singapore units of Alibaba, Baidu and Tencent, whose parents sit on a Pentagon list. It's legal, and it's reviving calls to control AI exports like chips.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
