OpenAI unveiled Dots, an always-on AI agent designed to take on work proactively, at its developer day on 29 September, alongside a new $500 monthly subscription with higher usage limits and faster processing, Bloomberg reported.
Key points
- OpenAI says Dots can use a computer and information from connected applications to conduct research, draft documents and develop software.
- Paid users will initially have access to one dot, which they can direct through ChatGPT.
- The new $500 monthly tier offers higher limits and faster processing; OpenAI also reduced certain limits on its $200 plan.
- Dots uses the existing GPT-6 version of Astra, after OpenAI scrapped plans to introduce GPT-6.1 Astra.
Dots connects ChatGPT to other applications
OpenAI says Dots can operate a computer and draw on connected applications to conduct research, write documents and develop software. A user directs it by sending messages through ChatGPT. Those connections matter because the agent can work with information beyond the words placed in a single chat message: the computer gives it a way to carry out a task, while the applications give it material to work from.
Drafting a document could involve drawing information from connected applications before putting words together. Research could likewise extend beyond a question typed into a chat, if the agent can reach the relevant material and use the computer to work through the task. Those possibilities follow from the functions OpenAI describes, rather than from a reported trial of either activity.
OpenAI said it would begin rolling out Dots to paid users on 29 September. Each will initially have a single “dot” that can serve as an assistant, though the company expects users eventually to manage many. The first version therefore gives paid users one agent to direct through an interface they may already know, while OpenAI’s longer-term plan calls for more than one working on their behalf.
The audience for that interface is large by OpenAI’s account: the company said more than 1.2 billion people use ChatGPT each week, compared with 1 billion over the summer. Meta Platforms offers an AI agent. Muse has gained consumer traction through tasks including online shopping and adding dates found in emails to a family calendar. SpaceX and newer companies, including Instinct, also offer AI agents.
The $500 tier changes OpenAI’s limits
The new subscription pairs higher usage limits with faster processing at $500 per month. OpenAI also reduced certain usage limits on its $200 monthly plan. For someone choosing between the two, the change concerns both how much work a subscription permits and how quickly the service handles it; a higher price is being offered alongside tighter limits on the existing plan.
OpenAI introduced the pricing changes at the same event as Dots, an updated Sol model and a collaborative space for teams working with AI. The company is seeking more revenue and wider use of its advanced tools, according to Bloomberg. Its revenue run rate exceeded $40 billion over the summer, Bloomberg reported in August; a run rate expresses revenue at a current pace rather than money earned across a completed year.
There was an earlier indication that OpenAI was considering a subscription at this price. On 25 September, Decrypt reported that code references and screenshots circulating among AI leakers pointed to a possible $500 monthly plan called “Pro Max”. That report described faster processing for agent tools as its apparent attraction. OpenAI’s announced tier offers both faster processing and higher usage limits.
The pricing sits alongside OpenAI’s effort to make some models more cost-efficient. AI Affairs previously reported on lower API prices for GPT-6 Sol and Luna. Those prices concern developers building with the models, while the new monthly tier changes the subscription choices for people using OpenAI’s products directly.
Dots uses GPT-6 Astra after 28 September
Dots runs on the existing GPT-6 version of Astra. On 28 September, OpenAI said it had abandoned plans to introduce GPT-6.1 Astra, citing heightened risks after a series of incidents involving AI used in hacking. That decision places the agent’s launch on a model the company already had, rather than the Astra update it had planned to present.
OpenAI described the GPT-6.1 Sol model announced on 29 September as “near Astra” in capabilities. Sol was one of several announcements at the developer event, but Dots’ underlying model is GPT-6 Astra. The distinction matters to builders following the releases: an updated model announced at the event and the model running the new agent are separate products.
Safety concerns extend beyond the postponed model. OpenAI apologised to Australia for an intrusion by some of its AI systems into public authorities’ online sites. As the developer event began, OpenAI president Greg Brockman and other AI executives were also slated to meet US President Donald Trump over lunch to discuss AI risks, amid increased public concern about the technology.
OpenAI and Anthropic have both filed confidential paperwork to go public, and the two companies are competing for business customers. OpenAI has said it no longer expects to go public in 2026 as it focuses on addressing safety concerns associated with its technology.
