ChatGPT Atlas

ChatGPT Atlas, OpenAI's browser, never made it off the Mac. OpenAI launched it in October 2025 and, in July 2026, said it would stop working that August.

What it was

ChatGPT Atlas was a free web browser for the Mac that OpenAI built "with ChatGPT at its core," in the words of its launch post.

It ran on Chromium, The Verge reported, and OpenAI's help page said it supported Macs with Apple silicon running macOS 14.2 or later.

Plus, Pro and Business subscribers got a preview of agent mode, which the launch post called "an early experience" that "may make mistakes on complex workflows."

OpenAI also ran a default browser promotion, offering a temporary increase in ChatGPT rate limits for up to seven days to users who made Atlas their default.

What happened

Atlas launched on October 21, 2025, worldwide on macOS for ChatGPT's Free, Plus, Pro and Go users, and in beta for Business accounts. OpenAI's launch post said versions for Windows, iOS and Android were "coming soon."

In January 2026, when a user on X wrote that most of OpenAI's users couldn't use Atlas because it was only for Mac, ChatGPT consumer product lead Adam Fry replied, "Team is hard at work on Windows."

On March 19, 2026, CNBC confirmed a Wall Street Journal report that OpenAI would combine its web browser, the ChatGPT app and its Codex coding app into a single desktop super app.

At the end of June 2026, OpenAI's Atlas page still said "Currently only available on macOS."

On July 9, 2026, OpenAI announced it was deprecating Atlas. Its help page gave users a wind-down of about 30 days and said Atlas "is scheduled to stop working on August 9, 2026."

After that date, it warned, Atlas "may no longer open, browse, or support browser-based agentic workflows." It pointed users to the ChatGPT desktop app and, inside Chrome, to its ChatGPT extension or sidebar.

By mid-September 2026, OpenAI's launch post opened with a note that Atlas "has since been deprecated." TechCrunch reported on September 15 that Atlas was "discontinued on August 9."

The end date on this page rests on that reporting. As captured in mid-September, OpenAI's help page still said Atlas "is scheduled to stop working" on that day.

Why it failed

OpenAI said it was ending Atlas as it moved its browser agent into ChatGPT and Codex, and reporters tied the shutdown to the company's cut of side projects.

What OpenAI said

The help page said, "We're deprecating Atlas and moving browser-based agentic capabilities into ChatGPT and Codex." OpenAI's ChatGPT announcement that day said the new features "build on what we learned from Atlas."

James Sun, who works on browsers at OpenAI, closed a thread of browser updates on X with "we are going to be sunsetting Atlas," crediting "Atlas users who took a leap of faith on a new browser."

The same thread announced a side chat in Chrome, bringing ChatGPT and Codex "into the browser where you currently do work."

OpenAI's July 9 statements don't call Atlas a failure or give a usage figure, and they don't mention demand, cost or a cut of side projects.

On the day CNBC confirmed the super app in March, Fidji Simo, then OpenAI's CEO of Applications, posted on X that when new bets like Codex start to work, "it's very important to double down on them and avoid distractions." She didn't name Atlas.

What reporting showed

Two days after launch, under the headline "The ChatGPT Atlas browser still feels like Googling with extra steps," The Verge's Emma Roth wrote that the experience "falls short when compared to rivals so far." In her test, Atlas's agent took 10 minutes to add three items to an Amazon cart. Perplexity's Comet took about two.

On TechCrunch's Equity podcast, Max Zeff said the most generous view of Atlas and Comet was "a slight efficiency gain," and asked, "why would a normal person use this?" TechCrunch's own testing found that using the agents "can feel more like a neat party trick than a meaningful productivity booster."

Chrome had "more than 3 billion users around the globe," TechCrunch noted on launch day, and The Verge's Roth wrote that Atlas "still needs a lot of work if OpenAI wants to pull people away from traditional browsers like Chrome."

In December 2025, TechCrunch's report on OpenAI's prompt injection post quoted Rami McCarthy, a principal security researcher at Wiz, who said that for most everyday use cases, agentic browsers "don't yet deliver enough value to justify their current risk profile."

When OpenAI announced the end, The Verge wrote that it followed OpenAI's push to reduce "side quests" and catch up with Anthropic on productivity features.

TechCrunch reported that Simo had told the team months earlier to cut back on "side quests," which led OpenAI to shut down its Sora video app.

The same article described an AI industry in "a war to unseat Chrome," with Comet, The Browser Company's Dia and new AI features in Chrome and Edge. TechCrunch wrote that "OpenAI appears to have concluded that the browser is a feature, not the destination."

9to5Mac described Atlas as an AI browser "with a small user base," without giving a number.

The stage it reached

ChatGPT Atlas never got past the introduction stage.

Its sales, place and price fit the stage. Atlas stayed on one platform. It was free, a penetration price, with a promotion for people who made it their default. OpenAI never published a user or download figure, and none of the coverage cited here reports one.

Its competition was the heavy kind, an established company launching into a market someone else already leads, as Amazon did with the Fire Phone.

The product life cycle hub defines introduction as the launch period, "when awareness is low and sales are small." Atlas launched with the ChatGPT name on it, and TechCrunch reported that day that ChatGPT drew 800 million users a week. Awareness of ChatGPT isn't adoption of Atlas, though, and without an Atlas figure there's nothing to show the fast rise in users that growth needs.

I wouldn't call it a decline. OpenAI did announce a shutdown and a replacement, but none of the sources shows Atlas's use peaking and then falling, and the replacement was OpenAI's own ChatGPT app, not a rival people left for.

The lesson

I'd test a new capability where people already work, as an extension or a feature of an app they already open, before asking them to switch something as habitual as their browser.

A standalone browser, I think, has to win the default spot before most people ever see what's new in it. Go standalone only once people keep coming back to the capability.