Dev & Open Source

OpenAI Codex Pricing in 2026: Every Plan vs Copilot, Cursor and Claude Code

OpenAI Codex pricing in 2026 runs from free to $200/mo across ChatGPT plans (Go $8, Plus $20, Pro 5x $100, Pro 20x $200), now metered by tokens. Here's every plan and how the cost compares to GitHub Copilot, Cursor, and Claude Code.

Waqas Ahmed Waseer
Waqas Ahmed Waseer Jul 18, 2026 8 min read
OpenAI Codex Pricing in 2026: Every Plan vs Copilot, Cursor and Claude Code

OpenAI Codex costs nothing to try and up to $200 a month at the top individual tier: it is bundled into every ChatGPT plan, from Free through Go ($8), Plus ($20), Pro 5x ($100) and Pro 20x ($200), with Business and Enterprise seats billed separately. Since April 2026 the paid tiers meter usage against API-style token consumption rather than a fixed message count, so what you actually pay depends far more on how hard you drive the agent than on the sticker price. If you are weighing Codex against GitHub Copilot, Cursor, or Claude Code, the headline number is only half the story. Here is what every Codex plan really costs in 2026, and how that stacks up against the rivals at the same price points.

What does OpenAI Codex cost in 2026?

Codex is not sold as a standalone product. It ships inside your ChatGPT subscription, and your plan sets the usage allowance. On the official Codex pricing page the tiers are Free ($0), Go ($8/mo), Plus ($20/mo), Pro at $100/mo for "5x" usage and $200/mo for "20x" usage, plus Business and Enterprise for teams. The ChatGPT developer pricing docs list the same ladder, with Business at roughly $20 per user per month on annual billing ($25 month-to-month) and Enterprise on custom pricing.

The catch is the word "usage." In April 2026 OpenAI switched Codex from per-message limits to token-based metering aligned with API token rates, applied to new and existing Plus, Pro, and Business plans. Each five-hour window carries a token allowance that scales with your tier, and heavier models (the GPT-5.6 Sol/Terra/Luna line) burn through it faster. That is why the same $20 Plus plan feels generous for one developer and cramped for another.

The real cost per developer

The subscription price is a floor, not a ceiling. OpenAI's own Codex rate card states that Codex costs roughly $100 to $200 per developer per month on average, with wide variance driven by the model chosen, how many parallel instances you run, and how much automation and "fast mode" work you push through it. In practice, a developer living inside an agent all day will exhaust a $20 Plus allowance quickly and land on Pro 5x or 20x.

There is a second path that most comparison pages skip: bring your own API key. Codex's CLI and IDE extension can authenticate with a raw OpenAI API key and bill per token at standard OpenAI API rates with no subscription at all. For light or bursty use this can undercut a monthly plan; for sustained daily use a flat Pro subscription is almost always cheaper than metered API billing. The decision mirrors the classic reserved-vs-on-demand tradeoff we see constantly in hosting: predictable heavy load wants a flat rate, spiky load wants pay-as-you-go.

Codex plans at a glance

PlanPrice (2026)Best for
Free$0Trying Codex on small, occasional tasks
Go$8/moLight individual coding, tight budgets
Plus$20/moOne developer, a few hours of agent work a day
Pro 5x$100/moDaily agent users hitting Plus limits
Pro 20x$200/moPower users running parallel agents all day
Business~$20/user/mo (annual)Teams needing admin controls and integrations
EnterpriseCustomLarge orgs, SSO, procurement
API key (BYO)Per-tokenLight, bursty, or programmatic use

How Codex pricing compares to Copilot, Cursor, and Claude Code

This is where the sticker prices get interesting, because the four leading tools cluster around the same $0 / $20 / $100 / $200 rungs but meter them very differently. Copilot leans on a monthly credit allowance, Cursor and Claude Code on usage pools that reset on a rolling window, and Codex on token metering. At the entry level, GitHub Copilot is the cheapest paid seat; at the ceiling, all four converge on $200 for their "power user" tier. The table below lines up the individual-developer plans so you can compare like for like.

ToolEntry paidMid tierPower tierMetering model
OpenAI CodexPlus $20/moPro 5x $100/moPro 20x $200/moToken-based (API-aligned)
GitHub CopilotPro $10/moPro+ $39/moMax $100/moMonthly credit allowance
CursorPro $20/moPro+ $60/moUltra $200/moUsage-credit pool
Claude CodePro $20/moMax 5x $100/moMax 20x $200/moRolling 5-hour + weekly limits

A few things stand out. GitHub Copilot is the only one with a genuinely cheap paid entry point at $10/mo (Pro), with Pro+ at $39 and Max at $100; its Business and Enterprise seats are billed per user. Cursor matches Codex on the $20 and $200 rungs but slots a $60 Pro+ in between, and its Teams plan is $40 per user per month. Claude Code is included in the same Claude Pro ($20), Max 5x ($100), and Max 20x ($200) subscriptions, drawing from a shared pool across web, desktop, and terminal — and it can also run on pure API billing, where Anthropic charges $5 per million input / $25 per million output tokens for Opus 4.8 and $3/$15 for Sonnet 4.6.

The honest takeaway: at $20 and $200 the four tools are priced almost identically, so the choice comes down to which agent and models you prefer, not the monthly fee. Codex's differentiator is the token-based meter, which makes cost predictable if you understand your token burn and punishing if you do not. For a deeper model-quality view, our best AI model for coding ranking and the Cursor vs Copilot breakdown cover the capability side that pricing alone can't.

Which Codex plan should you pick?

For most individual developers, Plus at $20/mo is the right starting point — enough for a few hours of agent-assisted work a day, and easy to upgrade if you hit the wall. Move to Pro 5x ($100) the moment you find yourself waiting out rate-limit windows more than once a week; that is the signal OpenAI designed the tier around. Pro 20x ($200) only pays off if you run multiple Codex instances in parallel or keep an agent working continuously through the day. Teams that need admin controls, shared billing, and GitHub or Slack integration should go straight to Business, which unlocks those features that Free and individual plans lack.

If your usage is genuinely light or spiky, skip the subscription and use an API key — you will pay only for the tokens you consume. We run this site's own publishing automation on agentic coding tools daily, and the pattern holds across every provider: match a flat plan to steady load and metered billing to bursty load. For picking the surrounding toolchain, our best AI IDE guide covers where Codex fits alongside the editor.

Frequently asked questions

How expensive is OpenAI Codex? Codex ranges from free to $200 per month for individuals. Paid tiers are Go ($8), Plus ($20), Pro 5x ($100), and Pro 20x ($200), with Business seats around $20 per user per month. OpenAI's rate card notes real-world spend averages $100 to $200 per developer per month once you account for heavy model use and parallel agents.

Is ChatGPT Codex free? Yes, there is a Free tier at $0 that lets you try Codex on small, occasional coding tasks through the web app, CLI, and IDE extension. It excludes cloud-based features like automatic code review and Slack integration, and its token allowance is limited, so serious daily use pushes you onto Plus or higher.

Is Codex cheaper than Claude? At the individual level they are priced almost identically: both offer $20, $100, and $200 tiers. Codex Plus is $20/mo and Claude Code is included in Claude Pro at $20/mo, with matching $100 and $200 power tiers. Which is "cheaper" depends on your token burn and preferred models, not the sticker price — on raw API billing, Anthropic's Opus 4.8 runs $5/$25 per million tokens.

Is OpenAI Codex worth it? For developers who use an AI coding agent daily, yes — $20 to $100 a month is trivial next to the time it saves. The value question is really about tier: don't pay for Pro 20x if Plus covers your usage, and consider an API key instead of a subscription if your use is light or irregular.

Sources

Some links may earn us a commission at no extra cost to you.

Waqas Ahmed Waseer

Waqas Ahmed Waseer

Waqas Ahmed Waseer is a developer and automation builder with 8+ years shipping production systems used by 100k+ people. He builds custom multi-tenant SaaS, AI automation (n8n, LLM workflows, WhatsApp bots) and hosting infrastructure (WHM/cPanel, CloudLinux) — and is the maker of WaSphere, FlowMaticX, and the WaseerHost hosting brand. 100+ projects delivered for SMBs, agencies and funded startups.

Related

More in Dev & Open Source

View all

Discussion · 0

Be kind. Comments are public.

    Newsletter · Monday edition

    The Monday brief.

    One email every Monday morning. The week ahead in AI, startups, hosting and dev tools — no fluff, no sponsored bait.

    Free. Unsubscribe in one click.