During the DevDay 2026 conference, OpenAI announced a package of more than 20 updates spanning ChatGPT subscriptions, developer tools, and enterprise services. Among them is the launch of ChatGPT Pro 500, the most expensive ChatGPT subscription to date, priced at $500 per month and providing access to the Astra Ultrafast model.
ChatGPT subscriptions expand beyond the app
The new subscription joins the existing Pro plans priced at $100 and $200 per month. After briefly pausing the acceptance of new subscribers to the Pro 200 tier, OpenAI reopened it with reduced usage limits, while eligible existing subscribers retain their previous limits until October 29, 2026.
The company also enabled the use of token capacity associated with ChatGPT subscriptions inside external tools through the Sign in with ChatGPT feature. The feature currently supports 16 partners, including Cognition’s Devin, Notion, Vercel, T3, OpenClaw, and Dactyl, with the ability to specify how much usage each tool consumes and count it against the subscription limits.
What is changing for developers?
Codex received updates that allow long coding tasks to run in the background without keeping the computer powered on. Codex Cloud environments enable developers to provide the system with approved repositories, dependencies, and permissions, then follow the work and review changes through the web or mobile and desktop applications.
Codex CLI adds voice commands and an /agents interface for tracking tasks that the user delegates to AI agents. Codex Security Cloud scans connected GitHub repositories, investigates findings, and monitors new commits. It can also suggest a code fix and create a draft pull request after reviewing it.
Agents operating within enterprise environments
The Agents API added support for computer use, allowing the creation of agents that interact with software and perform tasks on behalf of users. The updates include multi-agent capabilities in Codex, tool discovery and invocation, and conversation context compression, with OpenAI hosting the underlying infrastructure.
In partnership with AWS, OpenAI announced the initial preview of Amazon Bedrock Managed Agents, which combines OpenAI models with Codex and Amazon Bedrock AgentCore. The agents perform long-running tasks within AWS, while the runtime and inference environment remain inside it, and each agent is given an independent identity and defined permissions.
More targeted decisions and greater privacy
Decisions API provides Luna model capabilities for a predefined set of questions and answers, targeting classification tasks, request routing, and determining the agent’s next action using textual or visual context.
In workplace environments, it is now possible to mention @ChatGPT inside Slack and Microsoft Teams to use connected company tools, without requiring participants in enabled channels to have an individual ChatGPT license. OpenAI also announced a preview of Private Inference, which combines confidential computing and verifiable mechanisms for processing data, with plans to make it available during fall 2026.
Why does this news matter?
The DevDay 2026 announcements are not limited to raising model limits or adding a new interface; OpenAI is pushing to turn ChatGPT and Codex into an operating layer for agents inside developer tools and enterprise environments. In practice, this means shifting part of the work from conversation to executing long-running tasks, inspecting code, and interacting with software. However, the source does not specify pricing details or usage limits for each external tool, and Private Inference and Amazon Bedrock Managed Agents remain in the preview or planning stage, leaving the final availability scope unresolved.