OpenAI and Anthropic Launched Competing Super Apps on the Same Day

The ChatGPT Work vs Claude Cowork fight went from cold war to open war on Thursday. OpenAI launched ChatGPT Work, a super app that merges Codex into ChatGPT and produces documents, presentations, and hosted websites, powered by the brand new GPT-5.6 and pitched loudly as the cheaper option. Anthropic answered the same afternoon by expanding Claude Cowork from desktop to web and mobile, with remote sessions that keep working after you close your laptop and scheduled tasks that run with no device online. Both products chase the same idea: you describe an outcome, the AI works for hours, you get finished files. The two companies are racing toward IPOs and fighting over the same enterprise money. Regular people mostly can’t touch either yet, since both launches start on premium tiers.

Best for: understanding where AI tools are headed before your boss asks. Not ideal for: anyone hoping the chatbot era would last a little longer.

Thursday morning, OpenAI held a launch event. Thursday afternoon, Anthropic shipped a counterpunch.

Same day. Same product category. Same customers.

Nobody planned a truce, apparently. OpenAI unveiled ChatGPT Work, a so called super app that turns ChatGPT into a thing that produces finished work instead of finished sentences. Hours later, Anthropic expanded Claude Cowork to web and mobile, cutting the desktop cord on the product OpenAI had just spent a morning trying to catch.

The chatbot era didn’t get a farewell tour. It got a Thursday.

What OpenAI Actually Shipped

ChatGPT Work merges OpenAI’s chatbot with Codex, its AI coding tool, into one agent aimed at white collar workers. You describe what you need and it creates documents, presentations, and websites. Not website code you have to figure out what to do with. Actual hosted websites you can build and share directly through the app, thanks to a new hosted sites feature.

The launch details, reported by Reuters, sketch the rollout: web and mobile first, starting with Pro, Enterprise, and Edu users, expanding to Plus and Business over the following days. There’s a new ChatGPT desktop application shipping alongside it.

Under the hood sits GPT-5.6, which also debuted Thursday in three sizes: Sol at the top at $5 per million input tokens and $30 out, Terra in the middle at $2.50 and $15 (matching the old GPT-5.5 at half the cost), and Luna at the bottom at $1 and $6. The tiering is the strategy. Instead of one flagship price, OpenAI built an off ramp for every budget, then bolted its shiniest new product on top. That model has its own backstory: the launch was delayed last month at the US government’s request over national security concerns, part of the same regulatory season that took Claude’s Fable 5 offline for 19 days. Frontier AI now ships when Washington nods, and this week both labs got their nod.

The framing from OpenAI’s side was blunt. Product manager Ty Geri described the new model as competitive with far more expensive models at twice the speed. The launch pitch practically had a price tag stapled to it: coding tool power without the sticker shock.

Keep that phrase in mind. We’ll come back to whose sticker they meant.

Anthropic’s Same Day Counterpunch

Claude Cowork wasn’t new on Thursday. It launched in January, and we covered what Claude Cowork does when it landed: an agent that takes a goal, builds a plan, works through it, and hands you something finished while you’re off pretending you’ll get to your inbox.

The launch mattered. Fourteen million people paid attention to Cowork on day one back in January, and it turned the phrase computer use from a research demo into a product category. Dispatch followed, letting you fire tasks at your desktop from your phone. Cowork has been the thing OpenAI needed an answer to for six months.

What was new on Thursday: Cowork escaped the desktop.

Cowork now runs on web and mobile, with sessions that live in your Claude account instead of on your machine. Anthropic calls them remote sessions, currently in beta, and the practical meaning is the part that matters. Your work continues after you close your laptop. Scheduled tasks run with no device online at all. You can kick off a project from your phone at lunch, approve a step from the same phone an hour later, and open the finished result on your desktop that evening.

Chat and Cowork also now share one home, one place for projects and artifacts across both. The rollout starts with Max subscribers, who get doubled usage limits through August 5 as a welcome gift.

The timing reads exactly how it looks. OpenAI spent Thursday morning announcing it had caught up to January’s Cowork. Anthropic spent Thursday afternoon moving the target.

ChatGPT Work vs Claude Cowork: What’s Actually Different

Strip the launch theater away and the two products are converging on the same shape. Both take a goal instead of a prompt and run multi step work on their own. Each outputs real files: documents, spreadsheets, presentations, sites. And each exists because AI coding tools became more capable than chatbots, so both companies want non coders tapping that power without learning a terminal.

The differences live in the edges, and the edges are where you’d actually feel the ChatGPT Work vs Claude Cowork choice day to day.

OpenAI’s edge in the ChatGPT Work vs Claude Cowork matchup is distribution and price. ChatGPT has the largest user base in consumer AI, and Work reaches Plus subscribers within days. That puts an agent in front of tens of millions of people who’ve never heard the word Cowork.

Anthropic’s edge is maturity. Cowork has six months of real usage behind it, and Thursday’s update solved the complaint every agent product gets: what happens when I close the lid. Remote sessions and scheduled tasks are the difference between an assistant and an employee. One waits for you to show up. The other worked the night shift.

The Price Shot Was Aimed at One Company

OpenAI’s whole launch was wrapped in cost language. Cheaper. More broadly available. Without the sticker shock. Launch events don’t repeat a theme that hard by accident.

The context makes the target obvious. Two days before this launch, Anthropic’s flagship Fable 5 moved to usage credits at $10 per million input tokens and $50 per million output, on top of new ID verification for access. Premium capability, premium price, premium paperwork. Meanwhile the whole industry is mid tokenpocalypse: companies like Uber and Lindy have publicly cut AI spend, and enterprises are deploying prompt wrappers that force models into terse replies just to shrink the bills.

Into that anxiety, OpenAI walks on stage and says: ours is cheap.

It’s a real strategy, and it might work, but notice what it concedes. When your launch pitch is price, you’ve stopped arguing you’re better. Nobody markets a Ferrari by promising it’s affordable. OpenAI is positioning GPT-5.6 and Work as the sensible volume play while Anthropic sells the expensive top shelf, and both companies seem fine with those roles. There are worse businesses than being the affordable option in a market where the premium option requires a passport scan.

For what the models themselves trade off against each other, our ChatGPT vs Claude vs Gemini comparison covers the capability side of this fight in detail.

How the Two Handle the Same Job

Since most readers won’t touch both this month, here’s how the two flows are built, based on what each company shipped and documented.

Say the job is a competitor pricing report plus a simple landing page to share it. In ChatGPT Work, that’s one conversation. You describe the outcome. The agent researches, drafts the document, builds the deck if you want one, then generates and hosts the site through the new hosted sites feature. You get a link you can send to a colleague without ever leaving the app. The whole design bets on containment: everything happens inside ChatGPT, which is exactly what a super app means.

In Claude Cowork, the same job becomes a session. You set the goal, Cowork plans the steps, and you approve or redirect as it works through them. The difference shows up when life interrupts. Close the laptop mid report and the remote session keeps grinding on Anthropic’s side. Set the report to regenerate every Monday morning and the scheduled task runs whether any of your devices are awake or not. Approvals come to your phone.

Same destination, different philosophy. OpenAI built a place where work happens. Anthropic built a worker that doesn’t need you present. Which one fits depends on whether you want to watch the work or forget about it, and that’s the most honest way to frame the ChatGPT Work vs Claude Cowork decision until real hands on testing settles it (ours is coming once both reach the tiers we pay for).

The Fine Print Both Launches Share

Neither launch event dwelled on this part, so we will.

Agents make mistakes with more consequence than chatbots do. A chatbot’s error is a wrong paragraph you catch while reading. An agent’s error is a wrong number in slide 14 of a deck it built while you were at lunch. Your boss finds that one. Both products keep humans in the approval loop for a reason, and the reason is liability with your name on it. Review everything these tools produce like you’d review a new hire’s first month. A confident tone is not a citation.

Autonomy also widens the attack surface. Agents that browse, click, and execute can be manipulated by content they encounter along the way, a class of attack that’s grown up alongside the category all year. The more accounts and files an agent can reach, the more interesting you become to people who write malicious web pages for a living. Give these products access gradually, and treat every new permission like handing out a spare key.

None of that is a reason to skip the category. It’s a reason to start with low stakes work while the industry hardens things, because the industry is very much still hardening things.

Why Both Launched the Same Day

Nothing about Thursday was coincidence, and the reasons stack.

Both companies are preparing IPOs at valuations north of $800 billion, which means both are performing for the same future shareholders. Enterprise agents are the revenue story those valuations depend on. Selling $20 subscriptions to individuals is nice. Selling autonomous work to corporations replacing headcount is the actual business, and 2026’s 100,000 plus tech layoffs tell you how motivated those corporate buyers are.

There’s also a third player neither lab mentions by name. Microsoft has its own Copilot Cowork in the same category (yes, it borrowed the name, no, nobody’s embarrassed anymore), which means the productivity agent space now has three giants shipping into it simultaneously.

And hovering above all of it: the government. GPT-5.6 launched on a schedule shaped by the Commerce Department. Fable 5 came back online when export controls lifted on June 30, as Al Jazeera reported. A White House framework for voluntary frontier model standards is expected within weeks. Both labs cleared their regulatory runway within days of each other, so both fired within hours of each other. The launch calendar in AI now has two authors, and one of them works in Washington.

Which One Should Regular People Actually Use

Here’s the part every enterprise focused writeup skipped, and it’s the punchline.

Right now, most regular people can’t use either one.

ChatGPT Work starts with Pro at $200 a month, plus Enterprise and Edu accounts. Plus subscribers get it in the coming days, so the $20 tier is close. Claude Cowork’s web and mobile expansion starts with Max, which runs $100 to $200 a month, with other plans promised later. Both companies announced products for everyone and shipped them to the people already paying the most.

So the honest ChatGPT Work vs Claude Cowork answer today: if you’re a Plus subscriber, you’ll be trying ChatGPT Work before your Claude paying friends get Cowork on mobile, and that head start matters more than any feature chart. If you’re already on Claude Max, Thursday handed you the more complete product, and the doubled limits through August 5 are the moment to stress test it.

If you’re on the fence about paying anyone, our Claude Pro review breaks down what the $20 Claude tier actually gets you, and the calculus changes again whenever Cowork trickles down to Pro. Which it will. Both companies need these agents in cheap tiers eventually, because the entire point of the category is volume: millions of people delegating real work, not thousands of enthusiasts demoing it.

One practical note while you wait. The skills these products reward are identical: describing outcomes clearly, breaking work into checkable pieces, reviewing output skeptically. Every hour you spend getting good at that with the cheap tools transfers directly to the expensive ones.

What Thursday Actually Changed

The products will blur together within months. Features get copied in this industry faster than they get named, and by fall, hosted websites and remote sessions will be table stakes on both sides.

What doesn’t blur is the category shift both companies just confirmed. The chatbot, the thing you talk at and get paragraphs back from, is now the legacy product at both labs. The flagship is the agent: the thing that works while you don’t watch. OpenAI restructured its entire consumer app around that idea. Anthropic untethered its version from the desk. Neither company is hedging anymore.

For three years, the question people asked was which AI gives better answers. Thursday made the new question official: which AI does better work. Answers were a conversation. Work is a delegation. And delegation is the whole reason corporations are lining up with budgets while both labs sprint toward Wall Street.

The super app war didn’t start Thursday, really. It just stopped pretending to be about chat.