Summary
- OpenAI introduced "Dots," AI agents powered by GPT-6 Astra that operate on dedicated cloud computers, available for Pro and Business Premium users in select markets.
- According to OpenAI, GPT-6.1 Sol offers performance similar to GPT-6 Astra for coding and computer tasks at a fraction of the cost.
- The new Ultrafast tier boasts speeds up to 8 times faster in Codex and 6 times faster in the API, included in the $500-a-month Pro 500 plan.
At DevDay 2026, OpenAI emphasized the potential of AI as a collaborative worker equipped with its own computing power.
The company made over 20 announcements during the event in San Francisco, highlighting innovations ranging from the always-active agents called "Dots" to a confirmation of the anticipated $500 subscription tier and a collaborative workspace for humans and AI to edit documents together.
The event reinforced the concept of AI transitioning from merely a chatbot to an autonomous assistant capable of performing tasks and integrating with work applications even during downtime.
Just a day prior, OpenAI halted the release of its next flagship model, GPT-6.1 Astra, due to safety compliance issues. This decision follows incidents over the summer where OpenAI's agents interacted with unauthorized environments, including those of government entities and private firms.
Dots: Autonomous AI Agents
Dots are the centerpiece of this launch. OpenAI describes them as "remarkably capable, always-on agents" that take initiative rather than simply responding to queries and continue functioning even when the user is offline.
This concept parallels the efforts of open-source AI platforms such as Hermes and Openclaw, which have been exploring similar functionalities.
Each Dot operates on GPT-6 Astra, comes with its own cloud computer and browser, and can interact with over 4,000 applications, learning user preferences through feedback. Users can access their Dot via ChatGPT, voice calls, or platforms like Slack and Teams, with text messaging support on the way.
Dots adhere to specific protocols for when to act independently, ensuring that sensitive tasks, like password changes, remain under human control. The initial Dot is available to Pro and Business Premium users in select regions, while Free and Plus users do not have access.
ChatGPT Space: Collaborative Workspace
ChatGPT Space serves as a shared environment where users, ChatGPT, and Dots can collaborate on projects. It functions like a combination of a shared drive and group chat, allowing for the joint editing of documents. The introduction of "Pages" enables collaborative document creation.
During the keynote, OpenAI's Holly Li demonstrated how she assigned tasks to her Dot within a Space, such as drafting an FAQ and reviewing Slack messages. Collaborative slide presentations compatible with PowerPoint and Google Slides are expected to be available shortly.
GPT-6.1 Sol: Cost-Effective Performance
OpenAI's GPT-6.1 Sol is positioned as a new core model, launched shortly after GPT-6 Sol. OpenAI claims it nearly matches GPT-6 Astra in coding and professional tasks but at only one-fifth of Astra's token cost. Tokens are the units used for billing in AI services.
This announcement comes amid a challenging period for OpenAI, which recently postponed the October launch of GPT-6.1 Astra, with CEO Sam Altman indicating plans for a future release.
Sol is currently available via the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Educational users.
Ultrafast: Premium Speed Option
The Ultrafast tier is OpenAI's new high-speed offering, capable of generating tokens up to 8 times faster in Codex, reaching 300 tokens per second, provided users are willing to pay up to 6 times the standard rate. Speed is crucial as agents often perform numerous small tasks that require the model's assistance.
Astra Ultrafast is already operational, with a Sol version anticipated soon. It is offered on the Pro 500 and Enterprise plans.
However, a live demonstration of Ultrafast encountered several failures during the event, which was an unexpected way to showcase the speed advantage. This tier utilizes Cerebras chips, designed as large processors, since OpenAI's August preview.
Decisions API: Instant Decision-Making
The Decisions API directs Luna, OpenAI's most affordable model, to respond to questions with a set list of potential answers. Developers can submit text or images and receive back a selection to facilitate content sorting, request routing, or determining the next action for an agent.
This model functions more like a decision-maker than a thinker, emphasizing rapid responses to enhance reliability in API interactions. The API allows other software to communicate with the model and is currently in limited preview, with a broader rollout planned soon.
Codex Harness: Open Source for Developers
OpenAI announced that the Codex harness, which enables AI models to utilize tools and manage tasks, is now open source. This decision allows developers to examine and reuse the code that powers Dots.
OpenAI has made the Codex CLI, SDK, and app server available as open-source projects on GitHub, enabling external scrutiny of the agent's functionalities following previous incidents of agents breaching security.
Codex Security Cloud: Automated Bug Hunting
Codex Security Cloud utilizes OpenAI's coding agent for cybersecurity tasks, capable of scanning entire GitHub repositories on demand or on a schedule, reviewing new commits, identifying issues, and preparing fixes autonomously.
This feature integrates models introduced through Daybreak Blue, part of OpenAI's security initiative launched in May, and is accessible to Pro, Business, Enterprise, and Educational users.
OpenAI categorizes Astra as its first model with critical cyber capabilities, designed to detect and analyze unpatched vulnerabilities.
Agents API: Enhanced Developer Tools
The Agents API now enables AI to interact with software in a human-like manner, including clicking, typing, and navigating applications. It also incorporates Codex's multi-agent tools, tool search, and context compaction, which helps manage extensive task histories.
OpenAI manages the infrastructure, relieving developers of the burden of hosting. This API is integrated into Codex and ChatGPT Work on Pro 500 and Enterprise plans. Additionally, OpenAI announced Bedrock Managed Agents with Amazon, allowing OpenAI agents to operate entirely within AWS, although Plus and free users will not have access to this feature.
Private Intelligence: Data Security for Businesses
Private Intelligence is OpenAI’s solution for businesses wary of sharing sensitive information. It combines zero data retention—where the provider does not store user prompts or outputs—with automated safety checks conducted without OpenAI personnel viewing the data.
Upcoming is Private Inference, a preview set to launch this fall, which will execute model responses using confidential computing. This technology ensures the data remains secure within a locked environment, inaccessible even to the cloud operator, with verifiable controls. This offering is crucial for sectors like healthcare, finance, and legal services, differentiating itself from competitors like Anthropic, which retains some user data for extended periods.
Sign in with ChatGPT: Flexible Subscription Management
Typically, applications utilizing OpenAI models handle their own token payments. The "Sign in with ChatGPT" feature allows users to use their ChatGPT subscription across various tools, controlling how much each tool can consume.
OpenAI has established partnerships with 16 companies, including Cognition's Devin, Notion, Vercel, T3, OpenClaw, and Dactyl.
This feature is available for Plus and Pro users within participating applications, with global availability for the sign-in process. Usage will count toward the user's plan limits.
Plugin Extensions: Custom Integrations for ChatGPT
Plugin extensions enable external developers to create their own functionalities within ChatGPT, featuring a dedicated sidebar, interactive panels, and viewers for various file types supported by their products. OpenAI is expanding the platform used for its own features across all plans.
This initiative builds on previous launches, such as custom GPTs introduced at the first DevDay in 2023 and the transition of ChatGPT into an app platform in October 2025. These applications are now referred to as plugins, with a redesigned directory to rank and recommend them, including within conversations.
Users can select which plugins to activate and grant access permissions. With an estimated 1.2 billion weekly users, according to OpenAI, this presents a significant opportunity for developers.
OpenAI Marketplace: Allocate AI Budgets to Software
The OpenAI Marketplace allows qualifying enterprise customers to apply part of their existing OpenAI commitment towards approved partner software.
The initial 32 partners include Adobe and Figma for creative tools, Salesforce, HubSpot, and ServiceNow for customer service, and Harvey and Legora for legal applications, alongside security firms like Palo Alto Networks and CrowdStrike.
Enterprise clients can express interest in this feature now. Additionally, the Decisions API will be broadly available in the coming days, with GPT-6.1 Sol Ultrafast following shortly, and the Private Inference preview expected this fall.