The AI coding agent now runs in dedicated cloud containers, cutting task completion times by 90% and processing over 40 trillion tokens in its first weeks
OpenAI has turned its Codex platform into a full-blown cloud coding agent, giving developers the ability to offload programming tasks to sandboxed environments that run entirely independent of their local machines. The upgrade means writing features, squashing bugs, running tests, and even proposing pull requests can all happen in the cloud while a developer is, say, walking the dog or pretending to pay attention in a meeting.
The service is bundled into existing ChatGPT subscription tiers at no additional cost. Plus, Pro, Business, Edu, and Enterprise plan holders all get access, which is a notable pricing decision in a market where AI coding tools often come with their own separate bills.
From research preview to general availability #
Codex Cloud first appeared as a research preview on May 16, 2025, powered by the codex-1 model. Since then, OpenAI has steadily expanded its capabilities and pushed the platform to general availability with a suite of meaningful upgrades.
The headline performance number: a 90% reduction in median task completion time, achieved largely through container caching. In practical terms, that means the system provisions dedicated cloud containers pre-loaded with a user’s repositories. Instead of spinning up a fresh environment every time, Codex remembers the workspace and gets to work faster.
Daily usage has grown more than tenfold since early August 2025. And in its first three weeks after the GPT-5-powered version launched, Codex processed over 40 trillion tokens.
How the cloud architecture works #
Each Codex task runs inside an isolated, sandboxed cloud container. The container comes loaded with the relevant codebase, and the agent operates within that confined space without touching anything on the developer’s local machine.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The access points have multiplied as well. Codex Cloud is reachable through OpenAI’s web application, IDE extensions, GitHub integrations, Slack, and mobile devices.
A macOS desktop application for managing multiple Codex agents was introduced in February 2026, adding another layer of convenience for developers juggling several concurrent tasks. The app lets users monitor and coordinate multiple agent instances from a single dashboard.
Expanding the integration ecosystem #
OpenAI hasn’t kept Codex Cloud as a walled garden. The platform now integrates with AWS via Amazon Bedrock, which opens the door for enterprise teams already embedded in Amazon’s cloud ecosystem.
The GitHub integration is particularly interesting from a workflow perspective. Codex can propose pull requests directly, which means it doesn’t just write code in isolation. It participates in the collaborative review process that most modern development teams rely on. A developer can assign a bug fix to Codex, review the proposed changes in a pull request, and merge the fix, all without the agent ever touching the production branch directly.
Slack integration, meanwhile, targets the coordination layer. Teams can trigger Codex tasks and receive updates within the messaging tool they’re already using for daily communication.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our