OpenAI Dots: Cloud Agents Get Dedicated Browsers at DevDay

The company pairs persistent cloud environments with a dedicated browser for background tasks, while rolling out GPT-6.1 Sol and a $500 monthly plan.

At DevDay 2026 on 29 September in San Francisco, OpenAI introduced OpenAI Dots. The product moves software agents away from basic chat boxes and into background execution. Instead of keeping work trapped in a single window, each agent runs in its own cloud virtual machine with a dedicated web browser. Tech outlet The Information framed the launch as OpenAI’s fifth attempt to make autonomous agents work at scale, following earlier trials with Plugins, Custom GPTs, the Assistants API, and internal efforts like Operator.

What changes with OpenAI Dots

Standard assistants stop working the moment they finish typing an answer. Dots takes a different approach: the system keeps working even after you close your browser or step away from your desk. OpenAI says the service connects to more than 4,000 workplace applications, including Slack and Microsoft Teams.

Under the hood, Dots runs on the GPT-6 Astra model. Access rolled out first to paying Pro and Business Premium accounts. However, Pro subscribers in the European Union and the United Kingdom were excluded from this initial release because of local regulatory requirements. For Enterprise, Edu, and Healthcare customers, OpenAI made the tool available through an on-demand beta.

A history of attempts with agents

The count tracked by The Information points to a familiar pattern in OpenAI’s product history. In 2023, Plugins promised to turn ChatGPT into an extensible app platform, but poor usability and steep drop-off rates stalled adoption. Custom GPTs and the Assistants API followed, yet neither delivered the dependable automation developers needed for everyday corporate workflows.

Competition is stiffer now. Dots goes directly against Anthropic’s computer use features in Claude, alongside rival platforms from Meta and independent startups. To support this infrastructure, OpenAI updated its Agents API. The developer interface now includes computer use controls for screen navigation, native context compression, and multi-agent coordination managed on OpenAI’s own servers.

Developer tools: models, speed, and costs

DevDay also brought fresh pricing and performance options for builders. OpenAI introduced GPT-6.1 Sol as a budget model priced at one-fifth the token cost of Astra. For latency-sensitive systems, the company added an Ultrafast inference mode. OpenAI claims throughput of up to 300 tokens per second, roughly eight times baseline speed, charging six times the standard rate for the performance bump.

The company also introduced ChatGPT Space and an enterprise marketplace to distribute third-party tools across what it claims is a base of 1.2 billion weekly active users. For heavy users, OpenAI added the Pro 500 plan: a $500 monthly subscription offering 25 times the capacity of Plus, with Ultrafast access bundled in.

What remains unresolved

The pitch of round-the-clock agents still runs into clear operational limits. Dots currently requires explicit human confirmation before sending external messages or transmitting data outside the environment. In practice, that safety friction keeps unattended background tasks largely confined to reading and summarizing.

OpenAI has not set a release date for European subscribers kept waiting at launch. The company has also withheld revenue-sharing details for developers hoping to sell business tools in the new marketplace.

Sources

Press

Official

What changes with OpenAI Dots

A history of attempts with agents

Developer tools: models, speed, and costs

What remains unresolved

Sources

Sources

Markers