Codex
By OpenAI
Codex refers to OpenAI's line of AI coding models — originally a GPT-3 derivative released in 2021 that powered early GitHub Copilot, and relaunched in 2025 as a cloud-based coding agent built on newer GPT models.
Definition
Codex refers to OpenAI's line of AI coding models — originally a GPT-3 derivative released in 2021 that powered early GitHub Copilot, and relaunched in 2025 as a cloud-based coding agent built on newer GPT models.
Overview
OpenAI Codex was first introduced in 2021 as a version of GPT-3 fine-tuned on a large corpus of publicly available source code. It translated natural-language instructions into working code and was the model behind the original GitHub Copilot, one of the first widely used AI pair-programming tools. The original Codex API was eventually deprecated as its capabilities were absorbed into later general-purpose GPT models. OpenAI revived the Codex name in 2025 for a new software-engineering agent built on more recent models, designed to work semi-autonomously on coding tasks: reading a repository, writing and editing code, running tests, and proposing pull requests, going beyond simple code completion into agentic software engineering. This positions Codex alongside other AI coding agents as part of the broader shift from inline code suggestions toward tools that can carry out multi-step engineering work with AI agent style planning and tool use. Codex-derived capabilities are also used across OpenAI's developer tooling and integrate with the broader GPT model lineage, including GPT-4.1 and GPT-5, for tasks that mix reasoning with direct code execution.
Key Features
- Originated as a GPT-3 derivative fine-tuned on public source code (2021)
- Powered the original GitHub Copilot code-completion product
- Relaunched in 2025 as an autonomous coding agent
- Can read repositories, write code, run tests, and propose changes
- Supports natural-language-to-code translation across many languages
- Integrated into OpenAI's broader developer and agent tooling