Codex
FreemiumOpenAI's coding agent for the terminal, IDE, desktop, and cloud.
Published 29 May 2026 · Last updated 30 September 2026
Scores
Popularity4/5
67K GitHub stars on the open-source Codex CLI within its first year; reported 2M+ weekly active users across ChatGPT plans. Strong enterprise adoption but newer than incumbents like Cursor. Score 4.
Learning Curve3/5
Basic task creation via natural language is straightforward, but designing effective multi-step agent workflows, debugging sandbox failures, and managing parallel task coordination requires real practice. Score 3.
Flexibility3/5
Cloud-first architecture limits self-hosting; task environments can be customised via setup scripts and env vars, but core agent behaviour is proprietary and not deeply configurable. Score 3.
Performance5/5
Powered by codex-1 (o3-based), which leads SWE-bench and real-world coding benchmarks. Parallel execution eliminates sequential bottlenecks. Best-in-class for agentic coding tasks as of 2026. Score 5.
Portability2/5
Task execution is entirely cloud-bound on OpenAI infrastructure. Generated code and PR diffs are portable, but the agent itself cannot be self-hosted or migrated to another provider. Score 2.
About Codex
Codex is OpenAI's software engineering agent. You describe a task and it works on it end to end: it reads the repository, writes code, runs tests, fixes failures, and opens a pull request for review. It replaces the original Codex language model with a full agent system.
In the cloud, each task runs in its own isolated container with internet access off by default, and several agents can work on separate tasks in parallel. Locally, the same agent runs in your terminal, editor, or desktop app, where you can watch and steer it.
It is reachable in several places: the ChatGPT web app, an open-source Codex CLI (Apache-2.0, written largely in Rust, installed with npm, Homebrew, or a shell script), extensions for VS Code, JetBrains, and Xcode, a desktop app for macOS and Windows, and the ChatGPT mobile app. It also handles codebase questions and pull request review.
Codex Security adds vulnerability scanning and patch suggestions for repositories. Teams can delegate work asynchronously and spend their own time on design and review.
Codex is bundled into ChatGPT plans rather than sold on its own, so the plan sets the usage allowance, and extra usage can be bought with credits on paid plans. It can also run on an API key, where usage is billed at standard API rates.
Key Features
- End-to-end task execution: builds features, fixes bugs, runs tests, and opens pull requests
- Parallel agents on separate tasks, each in an isolated git worktree or cloud container
- Cloud sandboxes with internet access off by default
- Runs locally too, through the open-source CLI, desktop app, and IDE extensions
- Codebase Q&A and pull request review
- Available in ChatGPT web and mobile, CLI, VS Code, JetBrains, and Xcode
- Codex Security for vulnerability scanning and patch suggestions
Pros
- Hands-off delegation lets developers review results instead of driving each step
- Parallel tasks raise throughput on independent work
- Works well extending an existing codebase in its established patterns
- Pull request review catches edge cases that manual passes miss
- Open-source Apache-2.0 CLI can be inspected and customised
- Included in ChatGPT plans, so there is no separate subscription
Cons
- Cloud tasks run only on OpenAI's infrastructure, with no self-hosted option
- Limited visibility into the agent's reasoning makes its decisions harder to override
- Large refactors can still need several correction rounds
- Usage allowances are shared across ChatGPT features and cap heavy use on lower plans
- Vague prompts give poor results on complex tasks, so orchestrating it takes practice
Codex Pricing
Freemium- · Limited Codex access for quick coding tasks
- · Codex for lightweight coding tasks
- · Codex on web, CLI, IDE extension, and mobile
- · Enough for a few focused coding sessions each week
- · Extra usage through ChatGPT credits
- · Per user per month; $20 billed annually
- · Admin controls, SAML SSO, and MFA
- · No training on business data by default
- · About 5x the Codex usage of Plus
- · Extra usage through ChatGPT credits
- · About 20x the Codex usage of Plus
- · Extra usage through ChatGPT credits
- · Custom pricing through sales
- · SCIM, EKM, audit logs, and usage monitoring
Tech Stacks with Codex
Cursor Dev Stack
DeveloperCursor as both the editor and the coding agent, with Grok as the default model and Claude, OpenAI, or Gemini available from the same model picker, plus your pick of GitHub or GitLab for version control.
Devin Dev Stack
DeveloperDevin Desktop as the editor and agent manager, with Devin Cloud for work that keeps going without your machine, Claude as the default model and OpenAI, Gemini, or Grok in the same picker, plus your pick of GitHub or GitLab for version control.
Development
LLM
VS Code AI Dev Stack
DeveloperVS Code as the editor, with your choice of AI extension (GitHub Copilot by default, or Cline, Kilo Code, Claude Code, Codex, or Qwen Code) and your pick of GitHub or GitLab for version control.
Development
Learning Resources
View all →
Tools Related to Codex
Works well with Codex(8)
OpenHands is a self-hosted developer control center that can run and orchestrate Codex across local, remote, and cloud backends, alongside its own built-in agent and other ACP-compatible agents, per its own documentation.
Herdr gives Codex persistent, detachable terminal sessions and real-time status tracking, useful for long-running Codex tasks that need to survive a closed laptop lid or a dropped connection.
Codex CLI has no detach of its own — closing the terminal ends the run — so tmux is the standard way to keep a server session working across SSH disconnects.
Paseo gives Codex sessions cross-device continuity, letting a session started on a laptop be steered from a phone through Paseo's self-hosted daemon.
Conductor adds worktree isolation, a visual diff viewer, and GitHub/Linear integration on top of Codex, letting several Codex sessions run in parallel without conflicts.
T3 Code adds a desktop, web, and mobile control surface on top of Codex, letting a session be reviewed and steered without staying at the terminal it was started in.
Integrates with Codex(3)
OpenAI ships an official Codex extension for VS Code, allowing users to delegate tasks to the Codex cloud agent directly from the editor without switching to the web UI or CLI.
OpenAI's own coding agent, built and shipped by OpenAI to run its models.
Codex is one of the surfaces where OpenAI's Dots can be reached: a user's dot works inside Codex with the same context it carries in ChatGPT.
Alternatives to Codex(15)
Both are cloud/terminal AI coding agents by major AI labs; Codex (OpenAI) runs tasks asynchronously in cloud sandboxes, Claude Code (Anthropic) is a terminal-first agentic CLI — the closest alternatives in the agentic coding agent space.
Codex is a full-featured OpenAI agent across cloud, CLI, and IDEs; Pi is a deliberately minimal open-source terminal harness with four core tools, shaped through TypeScript extensions. Codex does more out of the box, Pi stays small and customizable.
Kiro (AWS) and Codex (OpenAI) are both AI coding agents backed by major cloud providers — Kiro integrates tightly with AWS services, Codex runs polyglot tasks in cloud sandboxes.
Codex is a cloud-based async coding agent; Cursor is a local AI-first IDE — different interaction models (asynchronous delegation vs. real-time pair programming), but both target developers who want AI-assisted coding.
Codex runs OpenAI models across cloud sandboxes, CLI, and desktop; ZCode is Z.ai's open-source desktop agent built around GLM. Codex adds cloud task runs, ZCode long local runs through Goal Mode.
Codex is OpenAI's cloud-based async coding agent; OpenCode is an open-source, model-agnostic terminal agent — alternatives in the agentic coding space with very different execution models (cloud sandbox delegation vs. local BYOK sessions).
Vendor
OpenAI
Website →Tags
Details
- Maintained
- Yes
- Primary languages
- JavaScript, TypeScript, Python, Go, Rust, Java
- Type
- AI IDE
- Open source
- No
- GitHub stars
- 126k
- AI features
- Chat, Agent
- Stars updated
- 2026-09-23