OpenAI Integrates Codex AI Coding Assistant Into ChatGPT

OpenAI announced on Friday the research preview launch of Codex, its most advanced AI coding agent to date.

Codex is driven by codex-1, a specialized version of the company's o3 AI reasoning model engineered for software development. According to OpenAI, codex-1 generates more refined code than o3, follows instructions with greater accuracy, and iteratively tests its code until all checks pass.

The Codex agent operates within a secure, cloud-based virtual computer. By integrating with GitHub, it can access your code repositories directly. OpenAI states this AI coding agent can complete tasks—such as writing simple features, fixing bugs, answering code-related questions, and executing tests—in anywhere from one minute to half an hour.

OpenAI notes that Codex can manage several software engineering tasks at once and doesn’t restrict users from using their computer or browser during operation.

Codex is available starting today for ChatGPT Pro, Enterprise, and Team subscribers. OpenAI indicates users will initially receive generous usage limits, but rate restrictions will be introduced in the coming weeks. After that, users can buy additional credits for Codex, an OpenAI representative confirmed to TechCrunch.

OpenAI also intends to extend Codex access to ChatGPT Plus and Edu users soon.

AI-assisted coding tools, sometimes referred to as "vibe coders," have gained significant traction recently. Executives at Google and Microsoft report that approximately 30% of their companies’ code is now AI-generated. In February, Anthropic released its Claude Code tool, and Google enhanced its Gemini Code Assist in April with more autonomous features.

Join us at TechCrunch Sessions: AI

Reserve your place at our premier AI industry event featuring experts from OpenAI, Anthropic, and Cohere. For a limited time, tickets are only $292 for a full day of talks, workshops, and high-impact networking.

Exhibit at TechCrunch Sessions: AI

Secure your space at TC Sessions: AI and present your technology to 1,200+ key decision-makers—without a large budget. Offer valid through May 9 or while tables remain.

The rise of AI-assisted coding has fueled rapid growth for the companies behind these platforms. Cursor, one of the most widely-used AI coding tools, achieved roughly $300 million in annualized revenue by April and is said to be raising new funding at a $9 billion valuation.

Now, OpenAI is entering the arena. The ChatGPT creator reportedly closed a $3 billion deal to acquire Windsurf, the team behind another leading AI coding platform. The introduction of Codex further demonstrates OpenAI's commitment to expanding its own AI coding capabilities.

Eligible users can find Codex in the ChatGPT sidebar. To assign a new coding task, type a prompt and select the “Code” button. Users may also ask questions about their code and click “Ask.” Below the prompt bar, users can track assigned tasks and monitor their status.

Before Codex's debut, OpenAI’s Agents Research Lead, Josh Tobin, explained to TechCrunch that the long-term goal is for these AI coding agents to serve as "virtual teammates," autonomously handling work that typically takes engineers hours or days. The company already uses Codex internally to automate repetitive jobs, build new feature scaffolds, and draft documentation.

OpenAI Product Lead Alexander Embiricos noted that many safety measures developed for the o3 model also apply to Codex. In a blog post, OpenAI confirms Codex will consistently deny requests to create harmful software. It also runs in an isolated environment without internet or external API access, reducing potential misuse—though this may also limit some functionality.

It's important to recognize that AI coding agents, like other generative AI systems, can make errors. A recent Microsoft study showed that top coding models like Claude 3.7 Sonnet and o3-mini had difficulty debugging reliably. Still, investor enthusiasm for these tools remains strong.

OpenAI is also upgrading Codex CLI, its open source coding agent for the terminal, with a version of the o4-mini model tailored for software engineering. This model is now the default in Codex CLI and accessible via OpenAI’s API at $1.50 per million input tokens (approximately 750,000 words) and $6 per million output tokens.

The introduction of Codex underscores OpenAI's strategy to enhance ChatGPT with specialized tools beyond its core chatbot. Over the past year, OpenAI has rolled out perks like priority access to Sora (its AI video generator), the Deep Research agent, and the web browsing agent Operator for subscribers.

These additions may attract new users to ChatGPT subscriptions and, in the case of Codex, encourage current subscribers to pay for higher usage limits.