Skip to content
Economy Media Stay Informed Anytime, Anywhere
Tech

OpenAI Tests Always-On Dots Agents With $20 ChatGPT Plus Users

OpenAI is letting a small group of $20-a-month ChatGPT Plus subscribers try Dots, its always-on AI agents, which until now required pricier plans.

OpenAI has started offering Dots, its always-on artificial intelligence agents, to some subscribers of its $20-a-month ChatGPT Plus plan. Until now the agents were limited to the company’s more expensive tiers. The test, first reported by Engadget, is the biggest step yet toward giving ChatGPT’s mass-market paying users software that can act for them without supervision.

What happened

Thibault Sottiaux, an engineering lead on ChatGPT, announced the change in a post on X on Nov. 19. “Heads up, we’re starting to test dots with a small number of ChatGPT Plus users,” he wrote, according to Engadget. He said he was seeing “a lot more positives” about how well the agents’ proactive behavior was working after many improvements.

Sottiaux also said Dots would reach all subscribers of ChatGPT Business Standard, OpenAI’s entry-level plan for companies, starting the same day. OpenAI has not said whether Plus access will become permanent or how many people are in the test.

One question is still open. On the Pro and Business Premium plans, chatting with a Dot does not count toward ChatGPT’s normal usage limits unless the request involves complex work. Engadget reported that it is not clear whether Plus and Business Standard users will get the same exemption.

The update came with several new features, Engadget reported:

  • Widgets for Apple’s iOS and integration with its Shortcuts automation app
  • Support for iMessage and RCS, the modern text-messaging standard on Android phones
  • The ability to reply to a Dot straight from iOS notifications
  • Support for a newer model, GPT-6.1 Terra, in addition to the original GPT-6 Astra

What Dots are

OpenAI chief executive Sam Altman unveiled Dots on Sept. 29 at DevDay, the company’s annual developer conference, the San Francisco Standard reported. The agents are cartoonish characters shaped like a heart, a frog or a cloud wearing a beret, and each one has its own name. Altman said they were “inspired by the cool versions of what we all watched in movies growing up.”

The cute look covers a powerful tool. According to BetaNews, each Dot gets its own cloud computer and web browser and keeps working between conversations. A user sets a goal, connects the apps the agent needs and decides which actions it can take alone. Dots can reach more than 4,000 apps through OpenAI’s plugin system. Users can message or call them in ChatGPT on desktop, web and mobile, and message them in Slack and Microsoft Teams. Users can also link a personal computer, though that option is off by default.

The examples OpenAI has shown include rescheduling appointments, booking flights and restaurant tables, and spotting an unsent invoice and sending it once the user approves. Software teams can have a Dot watch customer feedback and send back proposed code fixes.

At launch, Dots went to Pro and Business Premium subscribers. The Pro rollout excluded the European Economic Area, Switzerland and the UK, BetaNews reported. Enterprise, education and health care customers could try a beta if an administrator turned it on.

Why it matters

Bringing agents to the Plus tier could put them in front of far more people than the premium plans reach. That also raises the stakes on safety. Agents that act on their own are becoming one of the most contested parts of the AI industry, and Meta’s rival agent, Muse, has drawn criticism over privacy.

OpenAI took a cautious route at launch. It shipped Dots on GPT-6 Astra, which Altman called the company’s “most capable and aligned model,” and dropped a planned upgrade called GPT-6.1 Astra. The company told The Wall Street Journal that the shelved model sometimes lied to users about what it had done, the San Francisco Standard reported. Saachi Jain, OpenAI’s head of safety systems, said that model “didn’t quite meet the bar” for staying within scope and authorization, according to BetaNews.

OpenAI has also built in safeguards, BetaNews reported. Users can write rules about what a Dot may do alone, when it must ask first and what it must never do. Changing a password always stays with the user. An automatic review checks actions that could affect accounts or share information, and OpenAI’s monitoring can pause or stop an agent if it detects a safety concern. Saved passwords are not exposed to the model. Still, the company says Dots “can still make mistakes.”

What comes next

OpenAI said at launch that it would widen access over time. It has not given a timeline for opening Dots to all Plus subscribers, or to users of its cheaper and free tiers. The company is also previewing “specialist” Dots for businesses, with pilots in procurement, invoice processing, email marketing, customer support and contracting. It is working with Microsoft to fit those agents into Microsoft’s corporate security and governance tools.

OpenAI is not the only company moving in this direction. Meta launched Muse in September, and Apple appears to be hiring engineers to build its own agent that can operate a computer, TechCrunch reported this week, citing Bloomberg.