OpenAI has launched a new AI-first browser named ChatGPT Atlas.
OpenAI CEO Sam Altman said that AI gave them an opportunity to rethink how a browser should work in the AI era. ChatGPT Atlas seems to be the result of that initiative. “It’s a great browser. It’s smooth, it’s quick, it’s easy to use,” Altman said. The first version of ChatGPT Atlas is launching worldwide on macOS today to Free, Plus, Pro, and Go users, with more platforms and features likely to come soon.
The ChatGPT Atlas browser seems to be much like a regular browser — it has tabs and webpages — but has some crucial differences. It has a ChatGPT bar right at the center where the user can ask ChatGPT queries. They can also enter a URL in the bar to open up a new URL.
All URLs on the ChatGPT Atlas browser have an ‘Ask ChatGPT’ button on the top right. Clicking on the button opens a sidebar, through which users can chat with ChatGPT regarding the contents of the page. This can come in handy if the user wants to quickly summarize the contents of a webpage, or surface certain information from within it.
Interestingly, web searches default to ChatGPT. A seemingly specialized version of ChatGPT searches the web within Atlas. The results are similar to the results returned by traditional search engines, but are still within a ChatGPT-like interface. The search results have images, text links, and other media. As usual, users can click on the returned links to open these webpages.
The ChatGPT Atlas browser also allows users to write some text, for instance in an email, and right click on it with a quick instruction to ChatGPT. Users can write a rough email in their usual Gmail interface, but then right click, and write “tidy up this email”. ChatGPT immediately rewrites the bit of text in more formal language, all within the browser.
The browser also has a memory feature, which can be turned on or off. “If you turn on browser memories, ChatGPT can remember the pages you visit and use that context in helpful ways. For example, I look at a lot of gluten-free recipes but don’t always remember to make them. Now I can ask ChatGPT to make a meal plan for the week based on the latest recipes I’ve seen,” OpenAI’s Fidji Simo wrote in a blogpost.
ChatGPT Atlas agent
ChatGPT Atlas also has an agent. In the agent mode, users can write instructions on what they’d want the agent to do. Once they give the instructions through the sidebar, agents can take over, click, open new websites, and finish others tasks on behalf of the user.
The agent mode can also perform purchases on the user’s behalf. In a demo, ChatGPT employees went to a recipe webpage, and asked the agent to create a list of the ingredients required. From the list, they asked the agent to make purchases on their behalf. The agent browses to Instacart, searches for items, and creates a cart, and the user can finally make the purchase. OpenAI employees said that this feature would help users “vibe life”, as opposed to vibe coding, and get AI to finish their regular tasks.
The browser was highly anticipated — Google’s stock fell more than 3% in trade today when Sam Altman teased the news of the new browser. But after the livestream ended, Google’s stock has recovered most of its losses, and it trading close to parity.
It remains to be seen how much of an impact ChatGPT Atlas will have in the larger browser space. Many of its features, such as agent use, were already available through Perplexity’s Comet browser, which has been out for a few months. But ChatGPT is a much bigger brand, and will have many more opportunities to push its browser among users. But given the features in this first version, and given the friction involved in changing browsers, it does seem likely that ChatGPT Atlas, at least for now, will only be tested out by real ChatGPT power users and not the general public at large.