Codex

Codex is OpenAI's agentic software-engineering product: an AI agent that reads a repository, plans, edits files, runs commands and tests, and opens pull requests. It runs from an open-source terminal CLI, IDE extensions, desktop and web apps, a GitHub code-review integration and Slack, sharing one account and task history across all of them.

Codex is OpenAI's agentic software-engineering product: a coding agent that reads a repository, plans a change, edits files, runs commands and tests, and opens pull requests, rather than an autocomplete model. The name is reused from OpenAI's 2021 code model; the current product dates from the Codex CLI released on 16 April 2025 and the cloud agent powered by codex-1, an o3 derivative tuned for software engineering, in May 2025. It now runs across a deliberately wide set of surfaces that share one account and one task history: an Apache-2.0 terminal CLI written largely in Rust, which is one of the most-starred repositories on GitHub at roughly 106,000 stars; extensions for Visual Studio Code, JetBrains and Xcode; desktop applications for macOS, Windows and Linux; a web interface; a GitHub code-review integration; and a Slack integration. Model support has moved quickly, through GPT-5.3-Codex in February 2026, the roughly fifteen-times-faster GPT-5.3-Codex-Spark, and on to GPT-5.4 and GPT-5.5. For enterprise deployment, OpenAI has published its security architecture: execution is isolated in container-based sandboxes, outbound networking is default-deny with pre-approved endpoint allowlists, every suggested command passes an approval layer, and agent-native telemetry logs every action into queryable audit trails, with administrators able to allow or block plugins across teams. A dedicated Codex Security agent shipped in March 2026 for vulnerability hunting, reportedly surfacing nearly 800 critical and over 10,000 high-severity issues in testing across major projects. Adoption passed two million weekly active developers in March 2026, and in July 2026 OpenAI merged Codex into the ChatGPT desktop application, signalling a pivot from developer tool toward a general enterprise agent surface.

Category

Developer Tools

Pricing

Subscription, Usage-based, Freemium, Contact for pricing

Target Market

CTOs, VP Engineering, Enterprise Developers, Platform Engineers, Security Engineers

Deployment

Cloud-first, API-based, Open-source

Founded

2015

Headquarters

San Francisco, United States

Team Size

500+

Customers

2M+ weekly active developers (March 2026)

• ✓
  Multi-surface agent

  The same agent and task history run from CLI, VS Code, JetBrains, Xcode, macOS/Windows/Linux desktop apps, web, GitHub and Slack.

• ✓
  Open-source Apache-2.0 CLI

  The terminal client is Rust, public on GitHub at roughly 106,000 stars, so teams can inspect, build and pin it themselves.

• ✓
  Sandboxed execution with approvals

  Container-based isolation from the host plus a per-command approval layer, so no change reaches a repository without a human gate.

• ✓
  Default-deny network policy

  Outbound connections are blocked unless the endpoint is pre-approved, which limits data exfiltration and blast radius from a compromised dependency.

• ✓
  Agent-native telemetry

  Every action the agent takes is logged into queryable audit trails, giving the evidence trail compliance reviews actually ask for.

• ✓
  Codex Security agent

  A dedicated application-security agent shipped March 2026 that hunts vulnerabilities, reportedly finding nearly 800 critical issues in testing.

• ✓
  MCP plugin ecosystem

  Reported at over 9,000 Model Context Protocol plugins, the largest in agentic coding, with admin allowlists and private marketplaces for enterprises.

• •
  Delegated feature implementation

  Hand the agent a well-specified ticket and receive a pull request, with tests run in the sandbox before review.

• •
  Automated pull-request review

  The GitHub integration reviews incoming changes and flags issues, adding a consistent first-pass reviewer on every PR.

• •
  Infrastructure and CI/CD maintenance

  Terminal-shaped work such as pipeline fixes, IaC edits and dependency upgrades, which reviewers rate as its clearest strength.

• •
  Application-security triage

  The Codex Security agent scans repositories for vulnerabilities and ranks findings, reducing the manual sweep before a release.

• •
  Codebase exploration and onboarding

  New engineers ask the agent to trace how a subsystem works instead of reading through unfamiliar repositories unaided.

The Codex client itself is free and Apache-2.0; what you pay for is model access, and it is bundled into every ChatGPT plan rather than sold separately — Free $0, Go $8, Plus $20, Pro $100 for roughly 5x Plus usage or $200 for roughly 20x, Business per seat, Enterprise custom. Since 2 April 2026 OpenAI meters Plus, Pro and Business on token-based credits instead of per-message caps, with a typical task consuming roughly 5-45 credits depending on context size and mode, which makes forecasting harder than the old fixed limits. Business adds SSO, MFA, a dedicated workspace and a no-training-on-your-data guarantee; Enterprise replaces fixed limits with a shared credit pool and adds SCIM, RBAC and data-residency controls. Calling the Codex models directly through the API is billed separately per token. Rate limits remain the standing cost complaint: independent reviews report even $200/month Pro users hitting weekly caps on GPT-5.4.

This page was written from 6 sources, 6 on domains other than openai.com.

1. 1.github.com — codex
2. 2.learn.chatgpt.com — docs
3. 3.en.wikipedia.org — Codex (AI agent)
4. 4.codemyspec.com — codex cli review 2026
5. 5.eesel.ai — codex pricing
6. 6.artificialintelligenceherald.com — openai codex security architecture sandboxing telemetry 2026