OpenAI’s Codex costs $20 a month on the ChatGPT Plus plan, $100 or $200 a month on the two Pro tiers, and either $25 per seat (monthly) or usage-based credits for Business teams, a billing split OpenAI only introduced in April 2026. Used through the API directly instead of ChatGPT, Codex runs on a straight per-token rate that can cost more or less than a seat, depending on how heavily your team codes.
Key facts
- ChatGPT Plus ($20/month) includes Codex with a shared 5-hour usage window per model.
- Pro plans ($100 and $200/month) give roughly 5x and 20x that usage ceiling.
- Business workspaces can now choose usage-based Codex billing instead of a per-seat license.
- API access to the underlying models bills separately, by the token, with no monthly cap.
What each ChatGPT plan actually gets you
Codex access rides on top of your regular ChatGPT subscription, not a separate product you buy on its own. The limits differ by plan:
| Plan | Price | Codex usage |
|---|---|---|
| Free | $0/month | Minimal access, tight rolling window |
| Plus | $20/month | Standard 5-hour window, model-dependent message range |
| Pro 5x | $100/month | About 5 times the Plus ceiling |
| Pro 20x | $200/month | About 20 times the Plus ceiling, the highest individual cap |
| Business | $25/seat (monthly) or $20/seat (annual) | Plus-level limits per seat, plus a shared credit pool |
| Enterprise | Custom | Negotiated, usage scales with purchased credits |
Those “5-hour window” limits aren’t a fixed message count. OpenAI scales them by task complexity, so a quick code review question and a multi-file refactor draw down the window at very different rates, and the newer, heavier models (GPT-5.5 class) hit the ceiling faster than smaller ones like the mini tier.
Using Codex through the API instead
If your team wires Codex into CI, a custom agent, or an internal tool rather than clicking through ChatGPT, you’re billed per token on whichever model handles the request. As an example, the mid-tier GPT-5.4 model prices input tokens at $2.50 per million, cached input at $0.25 per million, and output at $15.00 per million, according to a rate breakdown from UI Bakery. Output costs six times more than fresh input, which is the part teams forget when they estimate from input volume alone.
That output-heavy pricing matters because Codex, like most coding agents, often returns far more tokens than it reads: a full file rewrite or a long diff. A task that reads 2,000 tokens of context and writes back 4,000 tokens of new code costs roughly $0.065 at those rates, not the few cents a naive “it’s just text” estimate would suggest.
The April 2026 shift nobody outside FinOps noticed
In April 2026, OpenAI moved ChatGPT Business customers away from a pure per-seat Codex license and toward usage-based billing, with free access enabled and charges tied to actual consumption, plus a $500 promotional credit per workspace, according to reporting at the time. Business Codex usage had reportedly grown sixfold since January, which is presumably why OpenAI wanted a billing model that scales with that growth instead of a flat per-seat number that doesn’t.
This is the opposite direction from GitHub Copilot and Cursor, both of which still sell straightforward per-seat subscriptions. A team split between light users and heavy daily users will usually come out ahead on usage-based pricing; a team where everyone codes with Codex all day might do better on a flat seat. There’s no universal answer, which is exactly why this is worth modeling with your own team’s actual usage before you pick a plan.
Keeping Codex spend under control
A few habits matter more than any pricing-page detail:
Route routine tasks, like small lint fixes or simple test generation, to a cheaper model tier rather than the top model by default; the price gap between a mini model and the flagship can be 6 to 10 times per token. Watch the 5-hour window usage per seat before it resets; a team that consistently maxes out Plus limits is often cheaper on Pro than on a pile of overage credits. And if you’re on Business, track actual workspace credit burn weekly rather than monthly. A spike that goes unnoticed for four weeks is a much bigger invoice than the same spike caught in week one.
FAQ
Is Codex included free with ChatGPT Plus?
Yes, Codex access is bundled into the $20/month Plus plan rather than sold separately, though usage is capped by a rolling 5-hour window.
Is Codex cheaper than GitHub Copilot or Cursor?
It depends on usage pattern more than list price. Copilot and Cursor both sell flat per-seat subscriptions; Codex now offers both a seat option and usage-based billing for Business teams, so a heavy user and a light user can land on very different effective costs. See our GitHub Copilot vs Cursor cost comparison for the seat-based side of that comparison.
Does Codex pricing include the underlying model API costs?
Only if you’re using Codex inside ChatGPT. Calling the same models directly through the API bills separately, per token, and that cost is on top of (not instead of) your ChatGPT subscription.
Sources
- JetAdmin: Codex pricing, current costs, usage limits, and realistic budgets (2026)
- UI Bakery: OpenAI Codex Pricing
- Pulse: OpenAI Shifts to Usage-Based Pricing for Codex in ChatGPT Business Plans
Written by the FinOpsDaily editorial team. Last updated: November 2026. Codex pricing changes frequently; confirm current plan prices and rate limits on OpenAI’s own Codex pricing page before budgeting.
