Not to let Claude steal the spotlight, OpenAI has also recently quietly launched a version that is far more significant than a “feature update”. Its AI programming assistant Codex desktop application, this time also adds three core capabilities:Computer Use、Built-in browser,as well asimage generation, and simultaneously released more than 111 new plug-ins. If you still think that Codex is just an auxiliary tool for automatically completing code, then you need to re-understand it.
OpenAI Codex big upgrade: can control your Mac, built-in browser, and generate images
Computer Use: Codex grows its own mouse cursor
The most eye-catching new feature of this update is that Codex has gained the ability to directly control any desktop application on macOS. It has an independent cursor, which can “see” the content on the screen, perform click actions, and enter text: Moreover, the entire process runs silently in the background and will not interfere with the user’s own operations. OpenAI particularly emphasizes the “background” feature: “Multiple agents can work in parallel without interfering with your operations in other applications.” This means that you can assign multiple Codex agents to handle different tasks separately while continuing to use your computer normally.
With computer use on macOS, Codex can now use any app by seeing, clicking, and typing with its own cursor.
It runs in the background without taking over your computer, working on tasks like frontend iteration, app testing, or any workflow that doesn’t expose an API. pic.twitter.com/iO9iubLZX9
— OpenAI (@OpenAI) April 16, 2026
In the official demonstration, users asked Codex to test a circle-crossing game (○× game) app: Codex automatically detected the project opened in Xcode, clicked the execution button to start the game, and started trial play. During the process, it discovered a bug that “the computer follows two steps at a time”, automatically corrected the code and re-executed it, and finally confirmed that the game can operate normally. The entire process requires no manual intervention.
Codex for (almost) everything.
It can now use apps on your Mac, connect to more of your tools, create images, learn from previous actions, remember how you like to work, and take on ongoing and repeatable tasks. pic.twitter.com/UEEsYBDYfo
— OpenAI (@OpenAI) April 16, 2026
This computer operation capability is closely related to what OpenAI acquired at the end of last year. Sky Applications(i.e. the original development team for Apple Shortcuts) are closely related. However, currently the Computer Use functionmacOS only, and is currently not open to EU and UK users.
Built-in browser Atlas: command directly on the web page
Codex now features its own browser technology based on OpenAI Atlas Built-in browser. This allows developers to directly specify specific sections of a web page without having to leave the Codex interface and switch to another tool. The actual application scenario is very straightforward: you can circle the slogan text on the web page and directly issue the command “reduce the font size and simplify the copywriting”, and Codex will automatically handle the corresponding front-end code modifications.
Currently this function mainly supports Local web application on localhost, OpenAI stated that it plans to further expand in the future to allow Codex to fully control the browser – including opening external websites, simulating user operation processes, screenshots and output inspection, etc.
gpt-image-1.5 integration: generate images in the same workflow
The third new feature is image generation, powered by OpenAI’s latest image model gpt-image-1.5 Provide support. Users can generate images directly in the Codex environment without switching to the ChatGPT application.
You can now generate and iterate on images with gpt-image-1.5 in Codex to create frontend designs, mockups, game assets, and more without leaving your workflow.
Usage is included with your ChatGPT account, no API key needed. pic.twitter.com/ay17I3Nxoa
— OpenAI (@OpenAI) April 16, 2026
OpenAI believes that this is most practical for front-end development. When establishing a product concept page or UI prototype, the design draft can be generated in the same workflow and then directly embedded into the corresponding page location, greatly reducing the conversion cost between design and development.
111 New Plugins: Connecting the Entire Tool Ecosystem
This update simultaneously releases more than 111 new Codex plugins, covering skill expansion, application integration and MCP (Model Context Protocol) server connection. Plug-in installation is extremely simple and can be completed in just a few clicks; enter the “@” symbol in the prompt word to call the specified plug-in.
We’ve also added support for 90+ plugins in Codex, giving it more ways to gather context and take action across the tools you already use for docs, project management, code review, creative work, deployments, and more. pic.twitter.com/IkmpDJwrLq
— OpenAI (@OpenAI) April 16, 2026
These plug-ins allow Codex to connect to a wider range of external tools and services, such as directly pulling issues on GitHub, integrating contextual information in Slack or Notion, and providing action suggestions based on this.
Scheduling and memory: From “command executor” to “autonomous collaborator”
In addition to the above three core new features, this update also strengthens Codex’s autonomy:
Scheduling functionCodex is now allowed to continue working on established conversation threads and schedule future tasks on its own. For example, in OpenAI, users can tell Codex “Monitor Slack, Gmail, Google Calendar, and Notion every hour to tell me important things.” Codex will automatically set periodic tasks to continue to execute. Long-term tasks spanning days or even weeks are also within the scope of support.
Memory function (Memory)This allows Codex to remember the user’s personal preferences, usual workflow, technology stack and other information, and proactively suggest work starting points based on the accumulated context – for example, reminding you of unanswered comments in Google Docs, sorting out relevant threads in Slack and Notion, and then giving you a prioritized action list. Currently, the memory function is not available to Enterprise Edition, Education Edition, and EU/UK users.
Development tool integration: GitHub, multi-terminal, SSH remote connection
In terms of development tool integration, this update adds:
- GitHub comment handling: Reply to the review comments on PR directly within Codex
- Multi-terminal paging: Open multiple terminals at the same time to execute different commands
- SSH remote connection (alpha): Connect to remote development environment
- Sidebar archive preview: Supports rich preview of PDF, spreadsheet, presentation and other formats
- summary panel: Track the agent’s plans, sources, and outputs
Subscription plan and user size
The update is on the way at the same time, and there is a new one Pro plan, priced at US$100 per month (approximately NT$3,250), which is between the existing Plus plan (US$20/month, approximately NT$650) and the high-end Pro plan (US$200/month, approximately NT$6,500). The new plan provides a usage limit that is 5 times higher than that of Plus, and will temporarily provide 10 times the usage limit of the Codex function during the initial launch period.
In terms of user base, OpenAI revealed that Codex currently has 3 million weekly active users, grew five times in three months, with a monthly growth rate of 70%.
Summary: From “AI assistant” to “digital worker”
The underlying logic of this Codex update is that OpenAI has quietly advanced its positioning from “a tool to help you write programs” to “a digital employee who can perform work on your behalf.” The addition of computer operation capabilities makes Codex no longer limited to the text world of the code editor; the integration of scheduling and memory functions gives it the ability to continue working across time and tasks.
In terms of competitors, Anthropic’s Claude Cowork has also supported PC operations, and AI agent control of the desktop environment is gradually becoming a new standard. The next question may no longer be “Can AI operate a computer?” but “Are we ready to let AI reside on our desktops?” This update has begun to be rolled out to Codex desktop users who log in with ChatGPT accounts.
Source: KOCPC Chinese