# OpenAI Codex

> Chat/LLM

- Canonical page: https://artifio.ai/models/openai-codex
- Run it: https://artifio.ai/create/openai-codex
- Modality: text
- Model brand: openai
- Pricing: from about $0.020 per generation (exact cost shown in the workspace before you confirm; varies with selected options)
- Typical generation time: ~30s (catalogue estimate)
- Full pricing table: https://artifio.ai/pricing/models

## About OpenAI Codex

Codex is OpenAI's coding agent line. The open-source CLI shipped in April 2025 and the cloud agent followed as a research preview on May 16, 2025; rather than autocomplete, it takes on whole software tasks in a sandboxed environment preloaded with your repository and returns finished work for review. The underlying models have moved fast: from codex-1, an o3-based reasoning model tuned for software engineering, to the GPT-5.x Codex line, with GPT-5.3-Codex arriving February 5, 2026, the low-latency Spark variant a week later, and GPT-5.4 powering Codex from March 5, 2026.

Growth matched the release pace: over 1 million developers by February 2026 and more than 2 million weekly active users by March. Artifio exposes the Codex model line as a text endpoint beside 100+ other models, paid per generation from one wallet.

## Features

- Runs each task in an isolated cloud environment with the target repository loaded, where it reads and edits files, runs tests, and invokes code-checking tools.
- Typical tasks complete in 1 to 30 minutes and return command logs and test results for inspection.
- Available as a web app, a CLI, Windows and macOS desktop apps, IDE extensions for Visual Studio Code, JetBrains, and Xcode, plus GitHub integration.
- The gpt-5.3-codex API model carries a 400,000-token context window, 128,000 max output tokens, and four selectable reasoning-effort levels.
- Codex Security, added March 2026, is an application-security agent that OpenAI tested against 1.2 million commits, flagging roughly 800 critical and over 10,000 high-severity issues.
- GPT-5.3-Codex-Spark is a roughly 15x faster variant for latency-sensitive work.

## What people use it for

- Large refactors and codebase migrations that span many files and need sustained context.
- Bug fixes delegated end to end: reproduce, patch, run the test suite, open a pull request.
- Parallel task delegation, farming several independent issues out to separate sandboxed runs.
- Security review of existing code with the Codex Security agent.
- Long agentic coding sessions from the terminal or an IDE.

## Reported strengths

- Fortune reported active weekly users tripled and weekly token volume grew fivefold after the GPT-5.3-Codex launch.
- Enterprise deployments include Cisco, Nvidia, Ramp, Rakuten, and Harvey.
- The sandbox-and-review loop keeps a human checkpoint between agent output and your main branch.
- OpenAI ships frontier coding models to Codex first, so the agent tracks the top of the company's coding stack.

## Reported limitations

- Output still needs human review; the product returns logs and test results precisely because unreviewed agent changes are not trusted.
- Cloud tasks run minutes, not seconds, with OpenAI citing a 1-to-30-minute range.
- Sandboxes originally had no internet access, and network access remains an opt-in rather than a default, which can block tasks that need to fetch dependencies.
- The market is crowded and moving quickly; Fortune notes rivals gained substantial enterprise share through early 2026 even as Codex grew.

## Sources

1. Codex (AI agent) (Wikipedia): https://en.wikipedia.org/wiki/OpenAI_Codex_(AI_agent)
2. GPT-5.3-Codex Model (OpenAI): https://developers.openai.com/api/docs/models/gpt-5.3-codex
3. OpenAI sees Codex users spike to 1.6 million, positions coding agent for enterprise (Fortune): https://fortune.com/2026/03/04/openai-codex-growth-enterprise-ai-agents/

---

About Artifio.ai: 100+ AI image, video, and audio models in one workspace with a single shared wallet — no per-model subscriptions. This is the markdown rendition of https://artifio.ai/models/openai-codex for AI assistants and crawlers; the HTML page is canonical.