[Review] Codex Free Tier Works, But the US$100 and US$200 Pro Pricing Never Showed Up on Four Official Pages

OpenAI’s Codex is a coding agent — software that can independently write code, fix bugs, and refactor files inside a cloud environment, rather than just suggesting code one line at a time. It ships through the command line (CLI), a web version, a desktop app, and plugins for VS Code, JetBrains, and Xcode. This review checks Codex’s free tier and pricing against what OpenAI actually publishes, because the two don’t fully match.

Codex isn’t sold on its own. It rides inside ChatGPT’s existing subscription tiers — Go, Plus, and Pro — and each tier hands you a different amount of Codex access. That bundling is the first thing worth understanding before anyone asks whether Codex pricing is worth it.

Free Access Exists, But It’s Labeled “Limited” for a Reason

Anyone with a free ChatGPT account can open Codex. No separate signup, no credit card. But OpenAI’s own account page marks free-tier Codex as “Limited” access — a real restriction, not a marketing footnote. The verified data doesn’t give an exact request count or daily cap for the free tier, so we won’t invent one. What’s confirmed: free access requires a ChatGPT account, Codex can be used for commercial work on the free tier, and there’s no watermark on anything it produces. If you run Codex through the CLI instead of the ChatGPT interface, you can plug in your own API key — in that case, you pay OpenAI directly for usage (API billing) instead of going through a subscription at all.

Three Paid Tiers, Three Different Amounts of Codex

Here’s where it gets confusing, because the pricing isn’t for Codex — it’s for ChatGPT, with Codex access changing by tier:

• ChatGPT Go, US$8/month — OpenAI’s own tier comparison labels this “Codex Limited.” You’re paying for Go’s other features; Codex itself stays restricted.
• ChatGPT Plus, US$20/month — labeled “Codex Yes” with expanded usage. This is the first tier where Codex access opens up fully rather than staying capped.
• ChatGPT Pro, US$100 or US$200/month — Pro itself is split into two price points. US$100/month reportedly gets you 5 times the Codex quota of Plus; US$200/month reportedly gets you 20 times the Plus quota.

That last point carries an asterisk, and it’s a big one.

Figure: compiled by AMPM from its verified data (last verified 2026-10-03).

The Pro Pricing Comes With a Sourcing Problem

When we rechecked this on 2026-08-11, all four official OpenAI pages we tried — openai.com/codex/, openai.com/chatgpt/pricing/, chatgpt.com/pricing/, and help.openai.com — returned HTTP 403 errors. A 403 means the request was blocked, not that the page doesn’t exist; it’s different from a 404, which would mean the content is actually gone. We retried through multiple paths and still couldn’t load an official page directly. The US$100-for-5x and US$200-for-20x figures come from cross-checking multiple independent third-party sources, not from reading OpenAI’s own page with our own eyes. Treat those two numbers as reported, not as independently confirmed against the primary source — and check OpenAI’s current announcement if the exact figure matters for a purchase decision.

🔍 AMPM exclusive check

This is the gap our research catches that OpenAI’s own marketing doesn’t spell out clearly: the word “pricing” on OpenAI’s site mostly describes ChatGPT, not Codex, and you have to dig into feature-comparison tables to see that Go, Plus, and Pro each hand you a different Codex access level — not a different price for Codex itself. OpenAI’s tier chart uses shorthand (“Codex Limited” vs. “Codex Yes”) rather than stating a request count or time window, which is why we can’t give readers a hard number for what “Limited” means on Go or on the free tier. The second gap is the one above: the US$100 and US$200 Pro figures, and the 5x/20x multipliers attached to them, could not be verified by us reading an official OpenAI page directly, because all four official entry points blocked our request with 403 errors during the 2026-08-11 check. We’ve rechecked the rest of this data on 2026-08-22, 2026-09-08, 2026-09-27, and again on 2026-10-03, and nothing has changed since — the free-tier description, the Go/Plus labels, and the Pro price points have stayed consistent across every recheck, just not independently confirmed at the source for the Pro figures.

☀️ AMO, the budget-minded cat

If you just want to try Codex without paying anything, the free ChatGPT tier gets you in the door — no card, no commitment. That’s genuinely useful for testing whether a coding agent fits how you work before spending a cent. But “Limited” is doing real work in that sentence, and OpenAI doesn’t publish an exact cap, so budget users should expect to hit a wall without knowing exactly when. The next step up, ChatGPT Go at US$8/month, is tempting because it’s the cheapest paid tier — except OpenAI’s own comparison chart still calls Codex “Limited” on Go. You’d be paying US$8 for Go’s other features, not for unrestricted Codex. The real unlock is Plus at US$20/month, where Codex access is marked “Yes” with expanded usage. For anyone who actually wants to use Codex as a daily tool rather than dabble, US$20/month is the realistic entry price, not US$8. I wouldn’t recommend Go to someone buying specifically for Codex — that US$8 doesn’t buy what people assume it buys. And I can’t tell you whether the US$100 or US$200 Pro tier is worth it for a budget user, because those numbers come from third-party sources, not OpenAI’s own page, and Pro’s price point is steep for anyone not already running Codex constantly.

🌙 PIMI, the performance-minded cat

What matters to me is whether Codex can actually carry real engineering work, and the access tiers map directly onto that. Free and Go both carry the “Limited” tag, which tells me not to build a workflow around either — you’ll get cut off mid-task. Plus at US$20/month is where Codex stops being limited and starts being usable for real: “Expanded usage” is OpenAI’s own phrase for it. For teams or individuals running Codex against production codebases all day, Pro’s two price points — US$100/month for a reported 5x Plus quota, US$200/month for a reported 20x Plus quota — are built for volume, not casual use. The 20x jump for US$200 only makes sense if you’re running Codex near-continuously across multiple projects; for anyone doing occasional refactors, that’s a lot of unused capacity. One flag worth repeating: the Pro numbers failed our direct verification against OpenAI’s own pages four separate times on the same day, blocked by 403s rather than confirmed absent. I’d treat the 5x/20x multipliers as directionally right but not gospel until OpenAI’s pages are readable again. The CLI’s pay-as-you-go API option is the one path here that sidesteps all of this — you pay for what you use, with no tier guessing involved, which is the option I’d point a performance-focused developer toward if the subscription tiers feel opaque.

Verdict: who should use it, who should skip it

Use the free tier to kick the tires — it costs nothing and needs only a ChatGPT account. Skip Go if Codex is your reason for paying; US$8/month still leaves you on “Limited” access per OpenAI’s own chart. Go straight to Plus at US$20/month if you want Codex to actually be useful day to day — that’s the tier where access opens up. Consider Pro’s US$100 or US$200/month only if you’re running Codex heavily enough to need 5x or 20x the Plus quota, and go in aware that those exact multipliers come from third-party confirmation, not a verified OpenAI page, as of our last check. Developers who’d rather avoid tier guesswork entirely can run the CLI against their own API key and pay by usage instead.

Illustration: Paying for the lower tier leaves the tool nozzle clamped and throttled, identical to the unpaid version.

Written: 2026-10-06 Price last verified: 2026-10-03 Tool last health-checked: 2026-10-06

🔗 Related on AMPM

• Codex pricing & free tier

📎 Sources

All prices and quotas in this article come from AMPM’s verified dataset for Codex (last verified 2026-10-03); primary sources:

• https://openai.com/codex/

🐾 Meet the cats: AMO and PIMI

AMPM is a Taiwan-based AI-tool pricing watchdog. Our motto: ask before you subscribe. Two cats argue every tool from two sides:

• ☀️ AMO — the budget-minded cat. Always asks: is the free tier enough, and is paying actually worth it?
• 🌙 PIMI — the performance-minded cat. Cares about whether it works well, fast, and gets your job done.

When they are done arguing, you know whether to pay. Every number is checked by us, with the verification date at the end.

🐾 Our sister sites also checked

• Travel eSIM plans compared
• Which credit card gives the best rewards on AI subscriptions