OpenAI Codex

OpenAI Codex has one of the more unusual histories in AI: the name first launched in August 2021 as a GPT-3 descendant fine-tuned on public code, powering the original version of GitHub Copilot and introducing the HumanEval benchmark, before OpenAI deprecated that model in March 2023. The name sat retired for two years until OpenAI revived it in April-May 2025 for something fundamentally different: an autonomous software engineering agent rather than a simple code-completion model. Codex CLI arrived first, followed a month later by a cloud-based agent that could enter a repository, make changes, run tests, and prepare a pull request for review, each task running in its own preloaded cloud sandbox.

Today's Codex is designed as a command center for agentic coding, built to drive real engineering work end to end, from routine pull requests to complex refactors and migrations, powered by OpenAI's frontier coding models including the GPT-5.6 family (Sol, Terra, and Luna tiers). Built-in worktrees and cloud environments let multiple Codex agents work in parallel across projects, and teams can use Skills to teach Codex their specific coding standards and workflows so it contributes more consistently with less supervision. Codex can also be scheduled for always-on background work like issue triage, alert monitoring, and CI/CD tasks, and its automated code review is designed to catch high-signal issues, including tricky backward-compatibility bugs, before they reach production. In March 2026, OpenAI added Codex Security, a dedicated agent for identifying and fixing software vulnerabilities.

Codex is the same agent everywhere a developer works: ChatGPT, a dedicated IDE extension, the terminal-based CLI, and the ChatGPT desktop app for macOS and Windows, all connected through a single ChatGPT account. By March 2026, Codex had grown to more than 2 million weekly active users, and companies including Duolingo, Cisco Meraki, Ramp, Harvey, and Sierra report substantial cuts to iteration time and expanded engineering capacity from adopting it. OpenAI positions Codex as evolving into a broader enterprise agent platform beyond just software development, with access bundled entirely into ChatGPT plans rather than sold as a separate subscription.

Codex has no standalone subscription; access is bundled into ChatGPT plans. Free ($0/month) includes limited trial access via a lighter Codex model with restricted daily limits. Go costs $8/month for light, local use only (no cloud task delegation). Plus costs $20/month and includes Codex on the web, CLI, IDE extension, and iOS, covering typical daily use. Pro splits into two tiers since April 9, 2026: Pro 5x at $100/month and Pro 20x at $200/month, offering 5x and 20x higher usage than Plus respectively. Business costs $20/user/month billed annually ($25/month billed monthly), with standard seats including Codex within usual plan limits; OpenAI stopped offering new pay-as-you-go Codex-only Business seats as of June 24, 2026, though existing seats continue working. Enterprise, Edu, and Gov plans use custom pricing. Since April 2, 2026, usage across Plus, Pro, and Business shifted from per-message limits to token-based credits (roughly $0.04 each), metered on a rolling 5-hour window plus a weekly cap; Enterprise, Edu, Health, and Gov plans moved to the same system on April 23, 2026. API-key usage bypasses ChatGPT plan credits entirely and bills directly at standard OpenAI API token rates. Real-world usage for active developers commonly runs $100 to $200 per month depending on model choice, parallel agents, and fast-mode usage.