OpenAI Unveils Codex, a Cloud-Based AI Assistant for Programmers

• OpenAI has launched Codex, a cloud-based AI agent.
• It writes code, runs tests, and creates requests to incorporate code changes on behalf of users.
• This helps accelerate the work of programmers in the cloud.

OpenAI has launched Codex, a cloud-based artificial intelligence (AI) agent for software engineers. It is already available for ChatGPT Pro, Team, and Enterprise users, and will soon be available for Plus users.

According to the official OpenAI blog, ”Codex is a software engineering agent that can work on multiple tasks in parallel: writing functions, answering code questions, finding and fixing bugs, and creating requests to incorporate code changes.”

Each task is performed in a separate environment in the cloud that is preloaded with the user's codebase. Codex can read and edit files, run tests, perform analyses, and conduct type checks-all without Internet access, which guarantees secure execution.

Codex is powered by codex-1, a special version of the OpenAI o3 model optimized for software engineering tasks. It has been trained using reinforcement learning on real engineering tasks to match human coding style and team standards.

Users can interact with the agent through a sidebar in ChatGPT, giving it tasks or asking questions about the code. Codex captures its actions through terminal logs and test results, allowing transparent verification of each stage of execution.

Early partners such as Cisco, Temporal, Superhuman, and Kodiak are already implementing Codex into their daily workflows.

Additionally, at Temporal, the agent runs complex tests in the background without distracting developers. At Superhuman, even product managers can make minor code changes without involving an engineer-just with validation.

OpenAI also released the lightweight codex-mini-latest model for Codex CLI, an agent that runs in the terminal. It responds faster to requests and is better suited for interactive code work.

Codex is available now at no additional cost during the exploratory preview phase. A flexible pricing model is planned for the future. For example, codex-mini-latest is available in the Responses API at $1.50 for 1 million input tokens and $6 for 1 million output tokens, with a 75% discount for repeat requests.