So what do I tell a non-coder, say someone running an Etsy shop who wants Codex tidying their order exports every week? Start with the free and then you’ll eventually upgrade because free is very limited. Do the free, then upgrade to twenty once you hit the pain point.

I’m on ChatGPT Pro 200 myself, and I run Codex on it most days. Has it ever stopped me there? No. And if it is, it’s because I’m going heavy as f* right now.

My real meter is further down.

Every ChatGPT plan’s Codex access, side by side

Prices come from ChatGPT’s pricing page, OpenAI’s Pro tiers article and OpenAI’s Business pricing, checked October 18, 2026. Codex access and windows come from OpenAI’s Codex pricing docs. Pro is billed monthly only.

Free and Go get Codex, but only Luna in the desktop app

OpenAI’s help center says it plainly: “Codex is included across ChatGPT plans, including Free and Go.” That’s per Using Codex with your ChatGPT plan.

What you get on those two is narrow. GPT-6 Luna at Standard speed, in the desktop app, subject to rollout. No Codex Cloud. No image generation on Free.

You also get limited ChatGPT Work on Free, and that matters more than it sounds. Work and Codex draw from the same meter, so a free account can feel the wall fast.

That’s fine. Free is for finding out whether Codex does anything useful for you at all.

Plus is the real Codex plan, and $20 goes further than you think

Honestly you get a lot for the twenty dollars. I think even now with the efficiency you get more with Sol.

Here is what OpenAI estimates for local messages per five hours on Plus. These are ranges, not promises:

That spread is why the model you pick matters more than the plan you pay for.

GPT-6.1 Sol became the default on paid plans when GPT-5.5 retired from Codex on October 14, 2026. OpenAI’s Codex models page calls it “near-Astra performance for complex work at a lower cost than Astra.” Run Sol by default and keep Astra for the hard stuff.

Weekly limits can apply on top of the five-hour window. Cloud tasks can use more of your allowance than local chats.

If you’re new to it, how to use Codex walks the first session.

Codex, ChatGPT Work, Excel and images all drain one meter

Regular ChatGPT chat has its own limits. Codex doesn’t get its own bucket.

OpenAI’s docs say it in five words: “ChatGPT Work and Codex share usage.” ChatGPT for Excel joins that pool on Plus and Pro. My own usage screen lists it as “Shared across Codex, Work, Workspace Agents, and ChatGPT for Excel. Chat conversations are not included.”

Some things burn it faster than a normal turn:

• Fast mode: 2.5x your included usage.
• Ultrafast on Astra or GPT-6.1 Sol: 8x. Among personal plans, only Pro 500 gets it.
• Image generation: about 3 to 5x faster on average, depending on size and quality.
• Voice in the desktop app: $0.05 per minute of your Codex budget.

Subagents count too. On OpenAI’s forum, coralcanvas614 wrote: “I got Pro 500, which now feels like Pro 100 or less.”

After checking, the same user found the cause: “turns out Astra spawned three Astra subagents.” That’s per the OpenAI community thread from October 18, 2026.

What my Pro 200 meter actually shows

This is my ChatGPT desktop app on October 18, 2026, Settings, then Usage & billing.

My ChatGPT usage screen on Pro 200, October 18, 2026.

Three things on that screen answer most Codex pricing questions:

• One meter, weekly only. No five-hour bar anywhere, which matches OpenAI’s “Pro plans currently have no five-hour limit.”
• 0 credits. I have never needed to buy any in the window below.
• Three banked full resets, expiring October 22, October 29 and November 6.

My Codex session logs back it up. Between October 8 and 18, 2026 they hold 4,138 rate-limit readings. Every one says plan Pro, every one shows a single weekly window, and not one says a limit was reached.

My Codex weekly meter from my own logs, October 8 to 18, 2026.

The week that started October 15 went from 0% to 84% used in about two and a half days. Then this happened.

Yesterday (October 17) we had a usage reset ahead of time. They announced it, so I was burning all of it. I mean, I already was.

I didn’t get to burn all of it. And then we got the reset, and I got more banked in.

My logs show the fresh week starting at 21:23 UTC on October 17, with the last reading before it at 84% used. My scheduled reset wasn’t due until October 22. So no, I haven’t had that issue.

What happens when you run out of Codex usage

OpenAI lets you finish the turn you’re in, “subject to fair use limits.” After that you have five ways forward, per OpenAI’s Codex plan help:

• Wait for the reset time shown in Settings, then Usage.
• Use a banked reset if you have one. It refreshes your five-hour and weekly windows and moves your weekly reset date.
• Buy credits on Plus or Pro. They last 12 months and work across Codex, Work, Word, Excel and PowerPoint.
• Buy an instant reset on Plus or Pro personal plans. It applies right away, pulls your normal weekly allowance forward, and is generally not refunded, per OpenAI’s paid reset article.
• Use an API key for extra local chats at standard API rates.

Support can’t reset your limits for you. OpenAI says so on the Pro tiers page.

OpenAI doesn’t publish a dollar price per credit. My screen shows “Up to 40% off” next to automatic reload, and the real price shows at checkout.

What do I do near the cap? Same as I said on October 2: “No, I don’t buy credits. I have two resets. So why would I? I’ll wait it out.”

Today it’s three resets, so the answer hasn’t changed.

When the $100 Pro plan is worth it for Codex

The twenty dollar plan will get you pretty far. Then you start building things you want to keep.

Once you start getting serious about building skills you’ll hit the roadblock. There’s just no way you’ll build a lot of skills and do a lot of brainstorming on $20 without hitting it.

That’s when you start considering the hundred dollar plan. Not just build skills, maintain them and build more, and run scheduled tasks on the back end, on the cloud, whatever.

One fact check on my own words: scheduled tasks and Codex Cloud are on Plus too. What $100 buys is the room to keep them all running every week without a five-hour wall.

Here’s what Pro adds for Codex, from OpenAI’s pages:

• No five-hour window. Just the weekly meter.
• More weekly room. OpenAI’s Pro tiers page publishes no multiple. OpenAI’s Tibo Sottiaux posted one on X in September, and I break that down in ChatGPT Pro $100 vs $200 vs $500.
• GPT-6 Astra Pro reasoning and longer Codex and ChatGPT Work sessions.

I’ve been on $100 and $200. In September I ran Codex on $100 with $0 in credits and one banked reset.

My ChatGPT usage screen on Pro $100, September 10, 2026.

After OpenAI shipped dots, its always-on agents, I went back to $200. They’re on Pro and not Plus.

merefield, on the same OpenAI forum thread, put the trade fairly: subscription accounts “are much less generous than before, but still represent great value compared to the alternative of using the API directly.”

Still torn between the two? ChatGPT Plus vs Pro runs the whole decision, not just the Codex part.

Codex with an API key costs per token instead

If you sign in with an API key instead of ChatGPT, you pay per token at OpenAI’s API prices. You lose cloud features like GitHub code review and Slack.

Those are standard rates for context under 272K, checked October 18, 2026. OpenAI warns not to use API prices to estimate what a subscription includes.

My rule for picking a Codex plan

• You’re curious and haven’t used it yet: Free. Find one job it does for you.
• You use it most weeks and keep hitting the wall: Plus at $20. Run GPT-6.1 Sol by default.
• You’re building skills, maintaining them, and running scheduled work every week: Pro at $100.
• You already live in Codex and Astra: look at $200, and watch one normal week before you decide.

Don’t pay for the wall before you hit it. Do the free, upgrade at the pain point, and let the meter tell you when it’s time for the next one.

If Claude Code is the other option on your list, Codex vs Claude Code is where I’d go next.

Where this comes from

The prices and limits are OpenAI’s, read off its own pages on October 18, 2026. What’s mine is my Pro 200 account, the usage screens, and the meter numbers from my Codex logs.

I haven’t run Codex on Free, Go, Business or Pro 500. Those rows are OpenAI’s word, not my test.

This post is part of Claude Guides, the hub for using AI without code.

Published October 18, 2026. Prices and limits checked against OpenAI’s own pages that day, linked inline.