OpenAI on Friday, Could 16, launched a brand new AI device referred to as Codex that’s designed to deal with a number of software program engineering-related duties on the similar time, from producing code for brand spanking new options to answering questions on a consumer’s codebase, fixing bugs, and suggesting pull requests for code evaluate
The cloud-based, AI agent-driven coding device runs these duties in its personal cloud sandbox atmosphere that has been preloaded with a consumer’s code repository.
Codex has been launched beneath analysis preview. Nevertheless, all ChatGPT Professional, Enterprise, and Crew customers have entry to the AI coding device. “Customers could have beneficiant entry at no extra value for the approaching weeks so you possibly can discover what Codex can do, after which we’ll roll out rate-limited entry and versatile pricing choices that allow you to buy extra utilization on-demand,” OpenAI stated in a weblog submit.
ChatGPT Plus and Edu prospects will probably be given entry at a later date, the Microsoft-backed AI startup added.
immediately we’re introducing codex.
it’s a software program engineering agent that runs within the cloud and does duties for you, like writing a brand new function of fixing a bug.
you possibly can run many duties in parallel.
— Sam Altman (@sama) May 16, 2025
https://platform.twitter.com/widgets.js
OpenAI’s newest providing comes at a time when AI is poised to disrupt the software program engineering sector, elevating widespread fears of job displacement. Microsoft CEO Satya Nadella not too long ago stated that 30 per cent of the corporate’s code is now AI-generated. A couple of weeks later, the tech big introduced it’s shedding 6,000 staff or 3 per cent of its workforce, with programmers reportedly being impacted essentially the most.
“It nonetheless stays important for customers to manually evaluate and validate all agent-generated code earlier than integration and execution,” OpenAI famous in its Codex announcement weblog submit.
What’s Codex?
With Codex, builders can delegate easy programming duties to an AI agent. It has its personal distinctive interface that may be accessed from the aspect bar within the ChatGPT internet app.
Story continues under this advert
Codex is powered by codex-1, an AI mannequin that may be a variation of OpenAI’s o3 reasoning mannequin. Besides that codex-1 has been particularly skilled on a variety of real-world coding duties to analyse and generate code “that intently mirrors human model and PR preferences, adheres exactly to directions.”
Its outputs have additional been fine-tuned utilizing reinforcement studying in order that codex-1 can “iteratively run checks till it receives a passing outcome.” When it comes to efficiency and accuracy, OpenAI stated that codex-1 fared higher than its o3 AI mannequin when evaluated on its inner SWE benchmark in addition to the firm’s human-validated model of it (SWE-bench Verified).
How does Codex work?
Codex can learn and edit information in addition to run instructions together with check harnesses, linters, and sort checkers. It usually takes wherever between one minute to half-hour to finish a process relying on the issue stage, as per OpenAI.
The AI coding agent performs every process in a definite, remoted atmosphere that’s preloaded with the consumer’s codebase serving as context. “Like human builders, Codex brokers carry out greatest when supplied with configured dev environments, dependable testing setups, and clear documentation,” OpenAI stated.
Story continues under this advert
Customers could make Codex work extra successfully for them by together with AGENTS.md information positioned inside their repository. “These are textual content information, akin to README.md, the place you possibly can inform Codex how one can navigate your codebase, which instructions to run for testing, and the way greatest to stick to your venture’s normal practices,” OpenAI additional stated.
One other distinctive function of Codex is that it reveals its pondering and work with each step because it goes about finishing the duty(s). Prior to now, a number of builders have identified that AI coding brokers produce coding scripts that don’t observe requirements and are troublesome to debug.
“Codex offers verifiable proof of its actions by means of citations of terminal logs and check outputs, permitting you to hint every step taken throughout process completion,” OpenAI stated.
As soon as Codex completes a process, it commits its adjustments in its atmosphere. Nevertheless, customers can even evaluate the outcomes, request additional revisions, open a GitHub pull request, or immediately make adjustments within the native improvement atmosphere.
Story continues under this advert
How you can use Codex? What are its use circumstances?
To ensure that Codex to begin producing code, customers must enter a immediate and click on on ‘code’. If they need the AI coding brokers to reply questions or present solutions, then customers want to pick the ‘ask’ possibility earlier than submitting the immediate.
When OpenAI opened up early entry to Codex for exterior companions, they used the AI coding agent device to speed up function improvement, debug points, write and execute checks, and refactor giant codebases. One other early tester used Codes to hurry up small however repetitive duties like enhancing check protection and fixing integration failures.”
It may also be used to write down debugging instruments and assist builders perceive unfamiliar elements of the codebase by surfacing related context and previous adjustments.
OpenAI builders are additionally utilizing Codex internally for refactoring, renaming, and writing checks in addition to scaffolding new options, wiring elements, fixing bugs, and drafting documentation.
Story continues under this advert
“Primarily based on learnings from early testers, we advocate assigning well-scoped duties to a number of brokers concurrently, and experimenting with several types of duties and prompts to discover the mannequin’s capabilities successfully,” the corporate stated.
What’s the distinction between Codex and Codex CLI?
In April this yr, OpenAI launched one other AI coding agent device referred to as Codex CLI. It’s stated to be an open-source, command-line device able to studying, modifying, and operating code domestically on a consumer’s terminal.
The coding agent integrates OpenAI’s fashions with the shopper’s command-line interface (CLI) used to run programmes, handle information, and extra.
Codex CLI is powered by OpenAI’s newest o4-mini mannequin by default. Nevertheless, customers can select their most well-liked OpenAI mannequin through the Responses API possibility. Codex CLI can solely run on macOS and Linux techniques for now, with help for Home windows nonetheless within the experimental stage.
Story continues under this advert
In Friday’s weblog submit, OpenAI additionally introduced updates to Codex CLI. A smaller model of codex-1 is coming to Codex CLI. “It’s accessible now because the default mannequin in Codex CLI and within the API as codex-mini-latest,” OpenAI stated.
The corporate has additionally simplified the developer log-in course of for Codex CLI. As an alternative of getting to manually generate and configure an API token, builders can now use their ChatGPT account to signal into Codex CLI and choose the API organisation they need to use. “Plus and Professional customers who check in to Codex CLI with ChatGPT can even start redeeming $5 and $50 in free API credit, respectively, later immediately for the following 30 days,” OpenAI stated.