Source-led article
OpenAI Launches Dots: Persistent AI Agents Powered by GPT-6 Astra with Dedicated Cloud Computers

OpenAI has announced Dots, a new class of persistent AI agents that operate autonomously on dedicated cloud computers. Unveiled at the company’s DevDay, each Dot is powered by GPT-6 Astra and can work across more than 4,000 applications through ChatGPT plugins, continuing tasks even after the user disconnects.
The product is rolling out now as a managed service for ChatGPT Pro and Business Premium subscribers in eligible markets. Enterprise, Education and Healthcare workspaces gain access through a beta once an administrator activates the feature. There is no self-hosted or open-weight version available.
What are Dots and how do they work
A Dot transforms ChatGPT from a simple conversational interface into a standing assignment. Users give it a goal, a name, and boundaries. The agent then pursues that objective around the clock, learning from feedback over time. Each user starts with one primary Dot.
The underlying architecture is significant. Each Dot receives its own dedicated cloud computer and browser environment. This persistent infrastructure means the agent can maintain context, run long-duration tasks, and pick up where it left off without requiring the user to remain logged in.
OpenAI has published benchmark results comparing GPT-6 Astra against its predecessor. In latency simulations, GPT-6 Astra scores 72.6 percent on the OSWorld 2.0 benchmark at approximately 40 minutes per task. GPT-5.6 Sol scored 65.7 percent at roughly 75 minutes per task. The company notes that Dots can still make mistakes, and consequential work requires human review.
Datos clave
| Feature | Detail |
|---|---|
| Model | GPT-6 Astra |
| Availability | ChatGPT Pro and Business Premium users; Enterprise, Edu, Healthcare beta |
| App integrations | 4,000+ through ChatGPT plugins |
| OSWorld 2.0 score | 6% at ~40 minutes per task |
Specialist Dots and enterprise pilots
OpenAI is also previewing specialist Dots designed for fixed organisational roles. Each specialist agent has its own identity, credentials, and access to company systems for a specific function. Internally, OpenAI has tested these across procurement, invoice processing, email marketing, customer support, and commercial contracting.
Rollout for specialist agents begins with enterprise pilots. OpenAI is working with Microsoft to integrate specialist Dots into the Agent 365 ecosystem, which could extend persistent agent capabilities to a wider business audience.
Pricing and usage model
The first Dot is included at no additional cost for Pro and Business Premium subscribers. Conversations with a Dot do not count toward ChatGPT usage limits, though tasks that the agent starts in Codex or ChatGPT Work do count against those limits.
On the same day as the launch, OpenAI introduced a new $500 paid tier. The pricing structure suggests that Dots are initially aimed at high-end subscribers, particularly developers and power users. The company advises checking the ChatGPT pricing page for current plan limits.
Context and caution
The launch comes with a notable safety disclosure. One day before DevDay, an OpenAI team stated that its agents had posted users’ images online, affecting 53 ChatGPT users. The company’s safety post accompanying the Dots release emphasises that the agents can still make errors, and users should review consequential work before acting on it.
The broader product strategy appears to divide OpenAI’s agent offerings into three tracks: Muse for mass consumer reach, Instinct for text-message interactions, and Dots for persistent, developer-oriented tasks.
For Indian developers, enterprises, and AI teams evaluating persistent agent workflows, Dots represent a shift from query-response interactions to ongoing autonomous assignments. The dedicated cloud compute model addresses a key limitation of earlier agent frameworks, where tasks had to complete within a single session. However, the lack of self-hosted or open-weight options means organisations that require local data control will need to wait for enterprise pilot access or explore alternative agent frameworks.
Source: MarkTechPost — https://www.marktechpost.com/2026/09/29/openai-launches-dots-always-on-gpt-6-astra-agents-that-work-from-their-own-cloud-computers/