Codex CLI pricing

OpenAI's open-source coding agent, bundled into ChatGPT plans or billed pay-as-you-go via the API

Prices verified against the official pricing page.

What you pay for Codex CLI depends on whether you’re running it against a ChatGPT subscription’s shared usage pool or against an API key billed by the token, and OpenAI keeps both paths open at once.

The subscription side stacks four tiers. Free costs nothing and lets you poke at Codex on quick local tasks, no cloud runs included. Go is $8/mo for lightweight work. Plus is $20/mo and is where most individual developers land: it includes Codex across the web, CLI, IDE extension, and iOS, plus cloud features like automatic GitHub code review and Slack integration, and it runs on GPT-6 Astra, added in early September, plus the GPT-5.6 family — Sol for depth, Terra as the everyday default, Luna for speed on high-volume work. Plus gives you 5-45 messages per 5-hour window on Astra, 10-100 on Sol, 25-200 on Terra, and 250-2,000 on Luna; Pro 5x and Pro 20x are exactly those bands multiplied by 5 and 20. Pro starts at $100/mo and buys 5x or 20x Plus’s rate limits depending on which Pro tier you pick, though OpenAI’s help center says new sign-ups and upgrades to the $200 20x tier have been paused since September 10, 2026. The research preview of GPT-5.3-Codex-Spark that Pro used to include was deprecated on September 14. Those bands are wide on purpose — usage is metered in credits that vary by model, context size, and how much tool-calling and retrieval a task triggers, so a five-line fix and a multi-file refactor can burn very different amounts of the same allowance.

Since late August OpenAI publishes the conversion, which makes the bands easier to reason about. Credits are charged per million tokens: 100 in / 10 cached / 500 out for GPT-5.6 Sol, 50 / 5 / 300 for Terra, and 5 / 0.5 / 30 for Luna, while GPT-6 Astra costs 250 / 25 / 1,250, two and a half times Sol, with the docs putting a typical GPT-5.6 message at 5 to 30 credits. That 20x spread between Sol and Luna is the real lever on how long your allowance lasts — bigger than the difference between two adjacent plan tiers. One caveat to note if you’re budgeting past the autumn: OpenAI describes Sol’s rate as promotional pricing available “at least through November 21, 2026.”

If subscription math doesn’t fit your workflow — CI runners, shared environments, or anyone who wants predictable per-task cost instead of a monthly ceiling — Codex CLI also runs against a bare API key. That skips the ChatGPT app entirely: no cloud tasks, no GitHub review, no Slack integration, just the CLI, SDK, or IDE extension billed at standard token rates. For gpt-5.3-codex, OpenAI’s coding-specialized model, that’s $1.75 per million input tokens, $0.175 per million cached input tokens, and $14 per million output tokens, confirmed directly on OpenAI’s API pricing page. Fast mode runs at exactly double those rates, $3.50/$0.35/$28 per Mtok; the page lists no Batch rate for this model. The output-token price is the number to watch: at $14/Mtok, a handful of long, code-heavy generations adds up fast compared to the input side.

Team and org pricing splits by seat rather than usage tier. Business is $20 per user per month billed annually (with a 2-seat minimum) or $25/user/mo billed monthly, and adds larger cloud VMs for faster task execution, SAML SSO, MFA, and a default no-training policy on your business data. Enterprise and Edu drop the fixed price for a sales conversation and layer on priority request processing, SCIM and EKM provisioning, domain verification, role-based access control, a compliance API with audit logs, and data residency controls — the checklist procurement teams ask for once Codex moves from a few developers experimenting to an org-wide rollout.

One structural detail worth knowing if you’re comparing Codex to Claude Code: OpenAI counts local CLI messages and cloud tasks against the same rolling 5-hour window, so running heavy cloud-based PR generation and interactive CLI sessions in the same afternoon draws from one shared pool, not two separate ones. Anthropic’s Claude Code, by contrast, sells usage as a flat multiple of Pro (5x on Max 5x, 20x on Max 20x) without splitting local versus cloud consumption the same way. If you’re deciding where Codex fits, that shared-pool design means heavy cloud automation (scheduled tasks, GitHub delegation) eats into the same budget as your interactive terminal sessions — plan Pro versus Plus accordingly rather than assuming cloud runs are “free” on top of local usage.

For most individual developers doing focused coding sessions a few times a week, Plus at $20/mo is the sensible starting point — it’s the tier OpenAI itself frames that way, and it includes the full surface area (CLI, IDE, cloud review) rather than Go’s stripped-down local-only access. Daily, heavy agentic users should expect to outgrow Plus’s rate limits and move to Pro, where the 5x-vs-20x choice comes down to how much of your day Codex actually runs rather than how much you’re willing to spend. While OpenAI’s pause on new 20x sign-ups lasts, though, 5x is the only one you can buy. And if you’re automating Codex inside CI or a shared service account where predictable per-token cost matters more than a flat monthly fee, the API key path — especially at gpt-5.3-codex’s $1.75/$14 rate card — is the more transparent option to budget against.

Plans

PlanPriceAnnualIncluded usageKey limits
Free Free — — Explore Codex on quick coding tasks · Local messages only, no cloud tasks
Go $8/mo — — Lightweight coding tasks · Codex on the web, CLI, IDE extension, and iOS
Plus $20/mo — — Codex on the web, CLI, IDE extension, and iOS · Cloud-based code review and Slack integration · GPT-6 Astra plus the GPT-5.6 family (Sol, Terra, Luna) · Per 5h window: 5-45 messages on GPT-6 Astra, 10-100 on GPT-5.6 Sol, 25-200 on Terra, 250-2,000 on Luna
Pro $100/mo — — 5x or 20x higher rate limits than Plus, priced per tier · Pro 5x: 25-225 Astra / 50-500 Sol / 125-1,000 Terra / 1,250-10,000 Luna messages per 5h · Pro 20x: 100-900 Astra / 200-2,000 Sol / 500-4,000 Terra / 5,000-40,000 Luna messages per 5h · Everything in Plus — Two tiers: $100/mo for Pro 5x and $200/mo for Pro 20x. The Codex plan card only says "From $100/month"; the $200 figure comes from OpenAI's help center ("Pro $200 unlocks 20x usage than Plus"), which also says new sign-ups and upgrades to Pro 20x have been paused since September 10, 2026. Existing Pro 20x subscriptions and Pro 5x are unaffected.
API Key Custom — — Codex in the CLI, SDK, or IDE extension · No cloud features (GitHub review, Slack) · Pay only for tokens used, at standard API rates — gpt-5.3-codex API pricing: $1.75/Mtok input, $0.175/Mtok cached input, $14/Mtok output (developers.openai.com/api/docs/pricing). Fast mode is $3.50/$0.35/$28 per Mtok; no Batch rate is listed for this model.
Business $25/user/mo ~$20/mo — 2+ users required for annual pricing · Larger cloud VMs for faster tasks · SAML SSO, MFA, admin controls · No training on business data by default — $20/user/mo billed annually (2+ users); $25/user/mo billed monthly.
Enterprise & Edu Custom — — Everything in Business · Priority request processing · SCIM, EKM, domain verification, RBAC · Compliance API, audit logs, data residency controls

Price change history

Frequently asked questions

How much does Codex CLI cost?

It's free to try, but real usage needs a ChatGPT plan or an API key. Go is $8/mo for light use, Plus is $20/mo, and Pro starts at $100/mo for 5x Plus's usage. If you'd rather pay per token instead of by subscription tier, an API key bills at standard rates — $1.75/Mtok input and $14/Mtok output for the coding-specialized gpt-5.3-codex model.

What's the difference between using Codex CLI on a ChatGPT plan versus an API key?

A ChatGPT plan (Go, Plus, Pro, Business) gives you a shared usage allowance across Codex's CLI, IDE extension, web app, and cloud tasks, refreshed on a rolling 5-hour window, plus cloud-only features like GitHub code review and Slack integration. An API key skips the subscription and cloud features entirely — you only get the CLI, SDK, or IDE extension, and you pay per token at OpenAI's published API rates instead of drawing from a plan's message allowance.