OpenAI rolled out a significant Codex update centered on GPT-5 Codex, a specialized version of GPT-5 tuned for agentic software work across terminals, IDEs, the web, GitHub, and the ChatGPT iOS app.
The company made GPT-5 Codex the default for cloud tasks and code review while keeping opt-in controls for local development via the Codex CLI and IDE extension, placing the agent inside existing developer workflows. Teams on ChatGPT Plus, Pro, Business, Edu, and Enterprise are in scope.
From helper to teammate, with endurance to match
OpenAI folded GPT-5 Codex into a unified Codex tied to ChatGPT, enabling seamless local-to-cloud handoffs and positioning it as a context-aware teammate for scoped tasks.
In tests, GPT-5 Codex operated autonomously for more than seven hours on difficult tasks, according to OpenAI. At the same time, it used 93.7% fewer tokens in the bottom decile and spent about twice as long reasoning in the top decile, pointing to speed on small jobs and endurance on large ones.
What’s new in GPT-5 Codex
OpenAI shipped a tuned model plus workflow upgrades across CLI, IDE, cloud, and GitHub review. Below are the key additions and how they work.
GPT-5 Codex model
The AI model was trained on real engineering tasks, building projects from scratch, adding features and tests, debugging, large refactors and code review. OpenAI said it is more steerable and adheres better to developer instructions.
Code review that targets critical issues
In evaluations on recent commits from popular open-source repositories, GPT-5 Codex produced fewer incorrect or unimportant comments while increasing high-impact findings, and validated behavior by executing code and tests.
Front-end and visual workflows
According to OpenAI, GPT-5 Codex improved human preference scores on mobile websites. In addition, when GPT-5 Codex is used in the cloud, OpenAI said it can inspect screenshots of its own work and attach progress images.
Codex CLI, rebuilt around agentic workflows
The open-source CLI was redesigned around agentic coding with image attachments for context, a to-do list for complex work, and connectors like web search and MCP. Approval modes were simplified to three levels, and long sessions can be compacted.
IDE extension for VS Code and Cursor
The extension brings the agent into VS Code and forks like Cursor. It shortens prompts by using open files and selections as context, and lets developers create, monitor, and complete cloud tasks without leaving the editor.
Cloud agent and GitHub integration
OpenAI said container caching cut median completion times for new tasks and followups by 90%. Codex can auto set up environments by detecting setup scripts and installing dependencies at runtime. On GitHub, it reviews PRs as they move from draft to ready and can be invoked with “@codex review”, including targeted checks such as security.
Security posture and safeguards
By default, Codex runs in a sandbox with network access disabled to reduce exfiltration and prompt-injection risks. Teams can restrict cloud network access to trusted domains and require approvals for risky commands in local tools.
OpenAI recommended using Codex as an additional reviewer, not a replacement for human review, and said GPT-5 Codex was treated as “High” capability in biological and chemical domains with added safeguards.
Rollout and availability
OpenAI set GPT-5 Codex as the default for cloud tasks and code reviews, with local use remaining opt-in via the CLI and IDE. Inside the AI company, Codex now reviews most pull requests and flags hundreds of issues a day.
The update is available across Plus, Pro, Business, Edu, and Enterprise plans; API access for CLI users is slated to follow, with Business credits and Enterprise shared pools to scale usage.