OpenAI used its annual developer conference in San Francisco on Tuesday, 29 September 2026, to make its biggest product push in years: more than 20 announcements centred on a single idea — that AI should stop being a chatbot you talk to and become a colleague that keeps working after you close the tab. The headliner was Dots, a new class of "always-on" AI agents, alongside a cheaper flagship-grade model, GPT-6.1 Sol, and a new $500-a-month subscription tier.
Chief executive Sam Altman delivered the keynote at Fort Mason, introducing what the company framed as the next phase of its product strategy: agents with ongoing responsibilities, shared workspaces where people and AI collaborate, and the infrastructure developers need to build their own agents at scale.
"Today, we introduced agents that can take on ongoing responsibilities and new ways for people and AI to work together."
— OpenAI, DevDay 2026 summary, reported by Thurrott.com
Dots: agents that never clock out
The centrepiece of the keynote was Dots — personal AI agents built into ChatGPT, powered by OpenAI's GPT-6 Astra model. Unlike a standard chatbot session, a Dot is designed to remain active between conversations: it gets its own cloud computer and browser, connects to more than 4,000 plug-ins, and keeps working on a goal in the background rather than stopping when the user closes the chat.
In practical terms, OpenAI described Dots as agents that can answer questions, explore ideas, provide feedback on projects, and carry out ongoing responsibilities across connected applications. Users can message their Dots inside ChatGPT, and — in a sign of how central OpenAI expects workplace software to be — also inside Slack and Microsoft Teams. Texting a Dot by SMS is coming soon, according to the company.
Availability is limited for now. Dots launched on 29 September for ChatGPT Pro 200 and Business Premium subscribers, with admin-approved access for Enterprise, Education and Healthcare plans. Free, Plus and standard Pro users do not yet have access, though OpenAI said more users will be onboarded over time.
The concept is not entirely new — rivals have been pushing in the same direction, with coverage of the keynote noting competing offerings such as Meta's Muse, SpaceX's Grok Bot and Google's Gemini Spark. But Dots represents OpenAI's most explicit bet yet that the future of its product is not better answers but delegated work: agents that plan, execute, monitor and continue without being prompted each time.
ChatGPT Space and Pages: the shared workspace
Closely tied to Dots were two collaboration products. ChatGPT Space is a shared workspace where teams — and their Dots — can work on projects together. Pages is a new document format designed for humans and agents to use side by side: collaborative documents that can incorporate research, images, charts and direct contributions from AI agents.
OpenAI also previewed collaborative slide presentations with AI assistance, and specialist enterprise Dots — company-configured agents for functions such as accounting, marketing and legal — with a planned Microsoft Agent 365 integration for managing them. The direction is clear: ChatGPT is being rebuilt as a shared surface where humans, agents and developers collaborate, reaching what OpenAI described as a collective audience of 1.2 billion weekly users.
GPT-6.1 Sol: near-flagship intelligence at a fifth of the price
For developers, the most consequential announcement may have been GPT-6.1 Sol — a new model just one week after its predecessor, GPT-6 Sol, rolled out. OpenAI described it as a major upgrade tuned for agentic coding, computer use and professional work, offering intelligence that nearly matches GPT-6 Astra on those workloads at one-fifth of Astra's standard input and output token prices.
The pricing details were striking. Cached input costs just $0.10 per million tokens — 95 per cent less than standard input pricing and 50 per cent less than GPT-6 Sol's cached input pricing, according to OpenAI. The explicit pitch: developers get more room to build and run capable agents that reuse context across requests, making it cheaper to execute complex workflows within the same budget.
GPT-6.1 Sol is available to all API users as well as Plus, Pro, Business, Enterprise and Education customers. It arrived on the same day OpenAI confirmed it had cancelled the launch of GPT-6.1 Astra — a sibling model — over safety concerns, underscoring that the company's model pipeline is now moving fast enough that some releases ship while others are held back.
Ultrafast and Pro 500: selling speed
OpenAI also introduced Ultrafast, a premium inference speed tier available in ChatGPT, Codex and its API. The company said Ultrafast generates tokens up to eight times faster in Codex and up to six times faster in the API — a feature aimed squarely at developers running agentic workloads where latency compounds across dozens of tool calls.
To go with it, OpenAI launched a new $500-per-month subscription plan — Pro 500 — with 25 times the usage allowance of Plus and access to Ultrafast. The company also reopened subscriptions to its existing Pro 200 plan. Astra is available on Ultrafast immediately, with Sol support to follow.
The pricing ladder now tells a story of its own: Plus for individuals, Pro tiers for power users and small teams, and a $500 tier aimed at the heaviest professional and developer workloads — an acknowledgement that AI usage, not seats, is becoming the unit of software pricing.
Codex grows up: cloud, security and voice
OpenAI's coding assistant, Codex, received some of the day's most practical upgrades. Codex in the cloud lets coding tasks keep running in a hosted environment after the developer closes their laptop — and lets users start or check on tasks from their phone. Reusable development environments and team-based settings and permissions round out the package.
A new Codex Security Cloud adds vulnerability scanning, investigation, deduplication and proposed fixes — positioning Codex not just as a code writer but as a participant in the security review process. The refreshed Codex command-line interface adds two-way voice interaction, a sign of how the company expects developers to work with agents conversationally, even in the terminal.
The Agents API: infrastructure for everyone else's agents
Perhaps the most strategically significant announcement was the public beta of the Agents API — managed agent infrastructure giving developers environments, sessions, tools, context management and multi-agent capabilities, with computer-use support so agents can interact directly with software. A Decisions API preview handles fast classification, routing and choosing agent actions.
OpenAI also announced Bedrock Managed Agents powered by OpenAI — agents running within Amazon Web Services — and "Private Intelligence," including private safety processing and a private inference preview, aimed at customers with strict data-handling requirements. Combined with Sign in with ChatGPT (launching with 16 partners), event-driven automations via new MCP event support, richer plug-in extensions, and an enterprise Marketplace, the message to developers was unmistakable: OpenAI wants to be the platform on which the agent economy is built.
What it all adds up to
Taken together, the DevDay announcements describe a company reorganising itself around a single thesis: the era of the chatbot is ending, and the era of the agent is beginning. Dots handle ongoing responsibilities. Space and Pages give humans and agents a shared surface. Sol makes capable agents cheap enough to run at scale. Ultrafast makes them fast enough to tolerate. The Agents API lets everyone else build the same pattern on OpenAI's infrastructure.
There are reasons for caution. Always-on agents that keep working in the background raise real questions about authorisation, oversight and data access — the same questions that led OpenAI to shelve GPT-6.1 Astra on the very same day it launched Dots. Enterprise buyers, in particular, will want to know exactly what a Dot can touch, what it reports back, and who is accountable when it goes beyond its brief.
But the direction of travel is now hard to dispute. With rivals fielding their own agent products and enterprise demand shifting from "a smarter assistant" to "work that gets done," OpenAI has placed its biggest bet yet on agents that don't wait to be asked. Whether users trust them enough to let them keep working while nobody is watching will decide how far that bet goes.
Sources
