OpenAI Codex Pricing: What the $20, $100, and $200 Plans Actually Get You
A breakdown of Codex subscription pricing on ChatGPT's $20, $100, and $200 plans, how usage limits reset, and when API billing makes sense.
What does Codex cost?
Codex, OpenAI’s coding agent, runs on the same subscription tiers as ChatGPT: $20 a month (Plus), $100 a month (Pro or a comparable business tier), and $200 a month (the top ChatGPT tier). There’s no separate Codex-only subscription. If you already pay for ChatGPT, Codex is included, and usage draws from a shared pool tied to your plan rather than a per-project or per-seat charge.
TL;DR
- Codex piggybacks on ChatGPT billing, so anyone paying $20, $100, or $200 a month for ChatGPT already has access, with no separate signup or extra fee.
- Usage is metered against a rolling allowance that appears as a percentage remaining inside the Codex app, and it resets on a weekly cycle rather than accumulating indefinitely.
- The $200 plan can translate into roughly $14,000 worth of inference if you use the full allowance, which makes the subscription far cheaper per token than paying for the same volume through API billing.
- API billing is available as a fallback once you exhaust your subscription usage, but it costs more per unit of usage than staying inside the subscription model.
- Free ChatGPT accounts don’t get meaningful Codex access, so anyone serious about using it should start at the $20 Plus tier and upgrade only if they hit the ceiling.
- The pricing model rewards heavy, consistent use, since the value comparison against hiring a person or paying hourly for equivalent output favors the subscription once usage climbs.
- ✕a coding agent
- ✕no-code
- ✕vibe coding
- ✕a faster Cursor
The one that tells the coding agents what to build.
How do the subscription tiers differ?
All three ChatGPT tiers enable Codex access, but the amount of usage scales with price. The $20 a month Plus plan gives the smallest allowance and is best suited to light, occasional use: quick coding tasks, small automations, or testing whether Codex fits your workflow. The $100 tier sits in the middle, giving more headroom for people running Codex daily across multiple projects. The $200 a month tier gives the largest allowance and is the one built for people using Codex as a daily driver, running long agent sessions, multiple projects, and repeated tool calls throughout the week.
The usage meter itself is visible inside the Codex desktop app as a simple percentage. A user on the $200 plan, for example, might see something like 94% of their weekly allowance remaining. That number ticks down as Codex performs work: reading files, running commands, searching the web, or executing multi-step agent loops. Once it hits zero, you either wait for the reset or pay for additional usage through the API.
How do usage limits reset?
Codex usage resets on a recurring cycle rather than a fixed monthly date tied to your billing cycle. In practice, that means if you burn through your allowance early in the week, you’re locked out of subscription-based usage until the next reset window rolls around, unless you switch to pay-as-you-go API billing to keep working.
This matters for planning. Someone running Codex heavily on a single sprint, like a several-day content push or a big automation build, can blow through a weekly allowance faster than someone spreading similar work evenly across the month. Builders who rely on Codex for daily operations should treat the usage meter the way they’d treat a data cap: check it periodically, and plan intensive work around where you sit in the cycle.
Is Codex worth it compared to API billing?
For most regular users, yes. The subscription pricing is structured so that the effective cost per unit of inference is meaningfully lower than paying for the same usage through OpenAI’s API. On the $200 a month plan, the usage allowance has been described as equivalent to roughly $14,000 worth of inference if fully used, a ratio that makes the subscription far more efficient than metered API billing for anyone using Codex regularly.
API billing still has a place. It works as an overflow option when you’ve exhausted your subscription allowance mid cycle and need to keep working without waiting for the reset. It’s also relevant for teams building Codex into automated pipelines where usage needs to scale beyond what any fixed subscription tier offers, or where billing needs to be tied to a specific project rather than a personal ChatGPT account.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
The practical framing that matters here: the subscription isn’t priced like a media or entertainment product. It’s priced against the alternative of hiring people to do the same work. At $200 a month, Codex is positioned as a productivity cost, not a discretionary one, and the comparison that matters isn’t “is this expensive for a subscription” but “what would it cost to get equivalent output any other way.”
Which plan should you start with?
Start at the $20 a month Plus tier if you’re new to Codex or unsure how much you’ll use it. It’s enough to explore the tool, run a project through its core workflows, and get a feel for how usage tracks against real work. If you find yourself hitting the usage ceiling regularly within a week, that’s the signal to upgrade to $100 or $200 rather than jumping straight to the top tier speculatively.
Free ChatGPT accounts aren’t a practical starting point for Codex. The paid tiers are where the actual usage allowance lives, so anyone planning to build with Codex, rather than just poke at it, needs at least the base paid plan.
One thing to watch: usage isn’t just about how many prompts you send. Codex’s agent loop involves multiple internal steps per request (reading files, reasoning, calling tools, responding, then repeating), so a single natural-language request can consume more of your allowance than the simplicity of the question suggests. Complex, multi-step tasks that involve browsing, file operations, or long reasoning chains will draw down the usage meter faster than a single short question.
Frequently Asked Questions
Does Codex have its own separate pricing plan?
No. Codex is bundled into existing ChatGPT subscription tiers ($20, $100, and $200 a month). There is no standalone Codex-only subscription; access depends entirely on which ChatGPT plan you’re already paying for.
What happens when I run out of Codex usage?
Once your weekly usage allowance is exhausted, you can either wait for the next reset cycle or switch to API billing, which charges per usage and costs more per unit than staying within the subscription allowance.
Is the $200 a month plan worth it for individuals?
For heavy, daily users, yes, based on the value comparison: a full allowance on that tier has been estimated as equivalent to roughly $14,000 of inference, making it far cheaper than paying the same volume through API billing. Light or occasional users are better served by the $20 or $100 tiers.
Can I use Codex on a free ChatGPT account?
Free accounts don’t provide meaningful Codex usage. Anyone planning to use Codex for real work should be on at least the $20 a month Plus tier.
How is usage measured inside Codex?
Codex displays a running percentage of your remaining allowance inside the app. That percentage decreases as the agent performs tasks such as file reads, tool calls, and multi-step reasoning, and it resets on a recurring weekly cycle.
