As a growing number of companies tap AI agents to automate software development tasks, OpenAI has refreshed its Codex coding agent with a focus on speed, reliability, and real-time collaboration across development platforms and infrastructure.
GPT-5 Codex aims to support two distinct developer needs: quick, interactive sessions and long, complex sessions involving a range of tasks. It works in various development environments: text-based Command Line Interface (CLI), the Integrated Development Environment (IDE) of choice, on the web, in GitHub, and in the ChatGPT iOS app.
OpenAI has unified GPT-5 Codex, first introduced in April, into a single product supporting these diverse development platforms so developers can work seamlessly between a local environment and the cloud.
OpenAI’s focus on developers should strike a chord: a recent report by Google on the ROI of AI found 40% of companies using agents in software development today, placing it among the top use cases for agents. And I certainly expect that figure to grow as developers and leaders including OpenAI continue to innovate with agents.
Core Features
GPT-5 Codex combines two essential skills for a coding agent: pairing with developers in interactive sessions, and persistent, independent execution on longer tasks including full-project engineering. Because it can reason across entire repositories, it can understand broader project context. GPT‑5-Codex now adapts how much time it spends on a task more dynamically based on the complexity of the task.
It will operate more quickly on small, well-defined requests or while a developer is interacting with it, and will work for longer periods on complex tasks like refactoring existing code for higher performance. The agent can accept input beyond text, including images and design diagrams, to assist with frontend development.
OpenAI noted that during testing, GPT-5 Codex has worked independently for more than seven hours at a time while it iterated, fixed test failures, and ultimately delivered code successfully. When compared to the general purpose GPT-5, GPT‑5-Codex uses 93.7% fewer tokens, a clear indicator of being fine-tuned for performance in a development context.
GPT‑5 Codex has been trained specifically for conducting code reviews and finding critical flaws. When reviewing, it navigates the codebase, reasons through dependencies, and runs tests to validate correctness.
AI Agent & Copilot Summit is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents. Building on its 2025 success, the 2026 event takes place March 17-19 in San Diego. Get more details.
Latest Enhancements
With new CLI enhancements, developers can now attach and share images — screenshots, wireframes, and diagrams — to build shared context on design decisions. In complex projects, the agent tracks progress with a to-do list and includes tools including web search and Model Context Protocol (MCP) for connecting to external systems.
Approval modes with CLI have been simplified to three levels: read-only with explicit approvals, auto with full workspace access but requiring approvals outside the workspace, and full access, with the ability to read files anywhere and run commands with network access.
GPT-5 Codex’s new IDE extension brings the Codex agent into VS Code, Cursor, and other VS Code variants, so developers can preview local changes and edit code with the agent.
It also lets developers move work between the cloud and local environments. They can create new cloud tasks, track in‑progress work, and review completed tasks without leaving their preferred editor.
The Codex cloud agent supports improved cloud infrastructure performance. By caching code containers, the completion time for new tasks and follow-ups has been reduced by 90%. The agent also now automatically sets up its own environment by scanning for common setup scripts and executing them.
New code review functions reason over an entire codebase and its dependencies, and execute code and tests to validate behavior. These features help teams find problems earlier, reduce reviewer workload, and deliver completed code with more confidence.
GPT-5 Codex is included with ChatGPT Plus, Pro, Business, and Enterprise plans from OpenAI.
Microsoft also announced that GPT-5 Codex is rolling out in public preview GitHub Copilot. It will be available to Copilot Pro, Pro+, Business, and Enterprise customers, who will be able to select the model in the Copilot Chat model picker. The company said its rollout will be “gradual.”
Ask Cloud Wars AI Agent about this analysis
