Guide
Estimated reading time: 8 minutes. Last verified: August 2026.
If you remember "Codex" as OpenAI's 2021 code-completion API (the one that originally powered early GitHub Copilot), that product was deprecated. What's called OpenAI Codex in 2026 is a different, agentic product: you describe a task, and Codex plans, writes, tests, and reports back, either locally in a terminal or asynchronously in a cloud sandbox against your connected GitHub repos.
The naming overlap with codex.ai, an unrelated third-party domain, adds to the confusion. If you're looking for OpenAI's coding agent, it lives at chatgpt.com/codex and inside ChatGPT's desktop app, not as a standalone product with its own pricing page.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Codex for light, occasional tasks; buy extra credits at the cap |
| Go | $8/month | Codex for lightweight tasks, higher limits than Free |
| Plus | $20/month | Codex across web, CLI, IDE extension, and iOS; roughly 15-90 local messages per 5-hour window depending on model |
| Pro (5x) | $100/month | 5x Plus rate limits |
| Pro (20x) | $200/month | 20x Plus rate limits, the highest individual tier |
| Business | $25/user/month ($20/user/month billed annually, 2-user minimum) | SAML SSO, MFA, no default training on your data, Codex via workspace credits |
| Enterprise | Custom | SCIM, EKM, RBAC, audit logs, data residency |
Sources: openai.com/chatgpt/pricing, chatgpt.com/codex/pricing. Verified Aug 1, 2026. Business replaced the previous $30/user/month Team plan on April 2, 2026.
There's also a fourth path: running Codex CLI locally with your own OpenAI API key, paying standard per-token API rates instead of a flat monthly fee. That's cheaper for occasional use and more expensive than Plus if you're coding with it daily.
Local messages and cloud tasks share a rolling 5-hour window, and the ceiling depends on which model you're running: the lighter GPT-5.6 Luna model allows roughly 50-280 messages per window on Plus, while the flagship GPT-5.6 Sol tops out around 15-90 in the same window. Pro's 5x tier roughly multiplies those numbers by five, and the 20x tier by twenty.
Practically: a developer doing a handful of focused Codex sessions per week fits comfortably on Plus. Someone running Codex as a daily background agent across multiple projects will hit Plus limits regularly and should budget for Pro from the start rather than upgrading reactively.
Not a fit for teams needing a fully air-gapped, on-premises coding assistant with zero cloud data flow, since cloud tasks and most plan tiers route through OpenAI's infrastructure.
For how Codex compares as an editor-embedded alternative, see Cursor vs Windsurf and GitHub Copilot vs Cursor earlier in this series.
Is OpenAI Codex the same as the old Codex API from 2021? No. The 2021 Codex API was a code-completion model, since deprecated. Today's Codex is a separate, agentic coding product bundled into ChatGPT plans, with a desktop app, CLI, and cloud sandbox.
Do I need a separate subscription for Codex? No. It's included in ChatGPT Free, Go, Plus, Pro, Business, and Enterprise plans. There's no Codex-only plan, though you can run Codex CLI against your own API key outside any ChatGPT plan.
What's the cheapest way to use Codex regularly? ChatGPT Plus at $20/month covers regular use. Occasional users may find running Codex CLI against a personal API key cheaper, paying only for tokens used.
Can Codex work entirely in the cloud without my machine running? Yes, cloud tasks run in a sandboxed environment against your connected GitHub repos and don't require your local machine or terminal to stay open.
2 curated tools below.
ChatGPT is OpenAI's flagship conversational AI on web, mobile, and desktop. It covers writing, coding, research, image/video, voice, and agentic work. As of late 2026, paid Work and Codex surfaces expose the GPT-6 family (including GPT-6.1 Sol), while Pro/Business Premium users in eligible markets get OpenAI Dots: always-on agents with their own cloud computer.
Gemini is Google's multimodal AI assistant and model family across the Gemini app, AI Mode in Search, AI Studio, and the Gemini API. As of October 2026 the frontier flagship is Gemini 4 Argon (paid API / Pro+Ultra surfaces); Flash-class models remain the cheaper workhorse for high-volume agents.