AI & TechArtificial IntelligenceBigTech CompaniesNewswireTechnology

OpenAI’s Codex Launches Free Mac App: Try It Now

Originally published on: February 4, 2026
▼ Summary

– OpenAI has launched a new Mac-only Codex app designed as a command center for managing multiple AI coding agents across long-term projects and tasks.
– The app features a security sandbox that limits folder writes and network access to make local AI operation safer for developers.
– Codex maintains context across different tools, allowing developers to switch seamlessly between the app, their IDE, and the terminal without losing their place.
– Usage of the latest GPT-5.2-Codex model has grown rapidly, with over a million developers using it recently and major enterprise customers adopting the tool.
– OpenAI is offering Codex for free on lower-tier plans for a limited time and is considering creating higher-speed, compute-intensive tiers for enterprise customers.

OpenAI has released a dedicated Mac application for its Codex AI coding assistant, marking a significant shift from its general-purpose ChatGPT app. This new tool is designed to function as a central command hub, enabling developers to manage multiple coding agents across complex, long-running projects. The launch highlights a strategic move to position Codex at the center of multi-agent software lifecycles, where programmers coordinate AI teams throughout design, building, shipping, and maintenance phases. The app aims to overcome interface limitations that users have encountered with previous models, particularly the highly capable GPT-5.2-Codex.

During a pre-launch briefing, OpenAI CEO Sam Altman emphasized the rapid evolution of software agents, noting that recent model advancements have unlocked new levels of practical utility. He described GPT-5.2 as a breakthrough, capable of handling extremely complex tasks, which led to the realization that the existing interfaces were holding users back. The new Mac app is the company’s answer to that problem.

Adoption of the latest Codex model has been swift. Since its mid-December debut, usage has nearly doubled, with over a million developers engaging with it in the past month. Altman called it the fastest-adopted model the company has ever released, with usage growing more than twentyfold since last August. Major enterprise clients, including Cisco, Duolingo, and Virgin Atlantic, are already on board, alongside a vast community of independent developers who report substantial productivity gains.

The application itself is built for parallel processing, allowing users to supervise several AI agents simultaneously on extended workflows. A key feature is the introduction of “skills,” which let developers create reusable automated routines for tasks like fetching logs, fixing tests, summarizing discussion threads, or completing tickets. While currently exclusive to macOS, a Windows version is anticipated to follow, mirroring the release pattern of the ChatGPT desktop app.

OpenAI also unveiled a new plan mode for Codex that offers a read-only review setting, preventing the AI from making direct code changes, and selectable agent personalities. Furthermore, the company recently extended Codex’s reach with an IDE extension for JetBrains environments, providing an alternative for developers who prefer those tools over the more AI-integrated VS Code.

A compelling aspect of the new app is its ability to maintain context across different development tools. During testing, the application seamlessly recognized and continued work from where a user last left off in their IDE, suggesting a fluid future where developers can switch between terminals, IDEs, and the Codex app without losing their place. Altman shared a personal anecdote, revealing he recently completed a substantial project over a few days without ever opening a traditional IDE, an experience he found astonishing.

Security on local machines is addressed through a controlled sandbox model. The app restricts folder write-access to user-approved directories, typically a project folder, and manages network permissions with remembered approvals over time. Developers can set approval levels ranging from “Untrusted” to “Never,” ensuring the AI operates within strict, user-defined boundaries.

Looking ahead, the OpenAI development team reports that their coding agents have become reliable enough to handle entire workflows within the new app. Several team members stated they’ve begun working almost exclusively in the Mac application, bypassing their IDEs and terminals altogether. For users, Codex is available with limited capacity on the $20 monthly ChatGPT Plus plan and with extensive resources on the $200 Pro plan. The company is also exploring even higher-tier options for enterprises that require maximum computation speed and can’t afford delays after major AI commands.

To spur adoption, OpenAI is offering Codex for free for a limited time with ChatGPT Free and Go plans, while doubling rate limits on Plus, Pro, Business, Enterprise, and Edu subscriptions. This opens the door for developers to experiment with coordinating AI agents for real feature development and bug hunting, potentially transforming how software is built and maintained.

(Source: ZDNET)

Topics

codex app 95% ai coding agents 93% gpt-5.2-codex 90% developer productivity 88% mac app 85% security sandbox 82% multi-agent workflows 80% ide integration 78% customer adoption 77% subscription plans 75%