OpenAI unveils "always-on" AI agents and a cheaper frontier model at DevDay 2026

OpenAI used its annual developer conference on Tuesday to push ChatGPT further from a chat window and into the role of a persistent digital employee, unveiling a new class of "always-on" agents called dots alongside a cheaper frontier language model and a workspace built for humans and software to edit documents side by side. The event, OpenAI DevDay 2026, was held at San Francisco's Fort Mason Center and opened with a keynote from chief executive Sam Altman, who framed the announcements as a shift from tools that answer questions to systems that keep working after the conversation ends.

The company teased more than 20 separate announcements over the course of the day, but the centerpiece was dots, described in OpenAI's own documentation as agents that operate continuously "across your tools and projects," maintaining their own memory of a user's preferences and ongoing work rather than starting fresh with every prompt.

What was announced

According to that documentation, a dot runs on a dedicated cloud computer with browser access, can be reached through ChatGPT's desktop and mobile apps as well as Slack, Microsoft Teams and voice calls, and is capable of pausing a task and resuming it later without a fixed schedule. The agents are powered by GPT-6 Astra, the company's top-tier model, and can connect to outside services such as Gmail, Google Drive and GitHub. Conversations with a dot do not count against a subscriber's ordinary ChatGPT usage limits, though the background tasks they run through OpenAI's Work and Codex products do.

Access is tiered. OpenAI's documentation lists dots as available to Pro subscribers on its 100, 200 and 500 dollar-a-month plans outside the European Economic Area, the United Kingdom and Switzerland, with a worldwide rollout for Business Premium customers and an enterprise version that administrators must switch on. The company describes the release as gradual, with no guarantee that any individual account will see it immediately.

Alongside dots, OpenAI introduced a collaborative hub it calls Space, which brings pages, spreadsheets, presentations, uploaded files and simple websites into a single shared location where a user and a dot can edit the same document, leave comments and pull in outside files from services like Google Drive while preserving existing sharing permissions.

The numbers

OpenAI also cut prices on its next tier of model down from its flagship. Per OpenAI's published model specifications, the new GPT-6.1 Sol model shares the same 1.05-million-token context window and 128,000-token output ceiling as GPT-6 Astra, and OpenAI describes it as delivering "near-Astra performance for complex work at a lower cost." The discount is steep: OpenAI's pricing page lists standard input tokens for Sol at $2 per million and output at $10 per million, compared with $10 and $50 per million for Astra — one-fifth the cost on both counts. OpenAI also introduced an "Ultrafast" processing mode that, according to demonstrations at the keynote reported by 9to5mac and Engadget's live coverage of the event, can generate roughly 300 tokens per second, several times the throughput of the standard model.

Beyond the headline products, OpenAI said usage of Plus and Pro subscriptions would extend into a small number of partner applications, including the AI coding tool Devin, while an enterprise marketplace would let corporate customers apply existing OpenAI spending commitments toward roughly three dozen approved outside vendors. The company also showed a new classification tool, referred to at the event as a Decisions API, intended to sort text and images quickly for developers building automated moderation or routing systems.

How OpenAI got here

The push into persistent agents builds on a strategy OpenAI has pursued unevenly since 2025, when it released an early autonomous browsing tool called Operator and later a general-purpose Agent Builder for developers. GPT-6 Astra, the model now powering dots, was itself released only weeks ago as OpenAI's most capable system, and the company has been racing rivals including Google's Gemini and Anthropic's Claude to move beyond single-turn chatbots toward software that can complete multi-step projects with minimal supervision. DevDay has become the venue where OpenAI translates that research into products aimed squarely at the developers and enterprise buyers who generate most of its revenue, rather than the general public that uses ChatGPT for free.

That commercial pressure has intensified as OpenAI's costs have grown alongside its user base. The company said during the keynote that ChatGPT now reaches roughly 1.2 billion weekly users, according to Engadget's live account of the presentation, a figure that underscores both the scale OpenAI has to defend and the incentive to move some of that traffic toward higher-margin subscription and enterprise products like dots and Space.

Who is affected

The most direct impact falls on the makers of workplace software that dots and Space are designed to sit inside or substitute for, including project-management and automation companies such as Notion, Asana and Zapier, whose own agent features now compete more directly with a tool built into ChatGPT itself. Enterprise customers who already pay for OpenAI's API face a pricing decision, since GPT-6.1 Sol offers most of Astra's capability at a fraction of the cost, while developers who build on OpenAI's platform gain a faster, cheaper option for applications that do not need the top-tier model. Millions of existing ChatGPT Plus and Pro subscribers, meanwhile, are the first to be offered dots as the feature rolls out gradually rather than all at once.

Reaction

Early hands-on accounts from people with pre-release access were mixed. Dan Shipper, chief executive of the media and software company Every, wrote in a review published the day of the event that the sheer volume of releases was unusual even by OpenAI's standards, calling it "a far greater number of new products than it has ever had" by his own count, while cautioning that the pace could outrun users' ability to absorb it.

"Dots has become the main way I use ChatGPT, but it's also been buggy," Shipper wrote, describing recurring permission and connection problems during testing, and adding: "Feature velocity is great when it feels coherent, but even a power user like me has a lot to figure out."

That tension between ambition and polish has followed OpenAI's agent products before, and the company's own documentation is candid that dots are rolling out gradually rather than as a finished, universally available feature.

What happens next

OpenAI said it would take the announcements on the road with DevDay Exchange events in eight cities, including Tokyo, Seoul, Paris, Berlin, London, São Paulo, Bengaluru and Mexico City, extending the pitch to developers who did not attend the San Francisco event. For now, access to dots and Space remains limited to the subscription tiers and regions OpenAI has specified, with wider availability dependent on how the initial rollout performs. The bigger test will play out over the coming months, as enterprise customers decide whether to route more of their work through OpenAI's own agents and workspace rather than the third-party software those products are designed to displace, and as competitors weigh their own responses in a race that increasingly centers on which company's software can be trusted to keep working unsupervised.