Codex CLI pricing for startup teams: seats, API keys and what to measure

Codex CLI pricing comes down to one point: the tool costs nothing, and you pay for the model time behind it, either through a ChatGPT plan or through an OpenAI API key. As of November 2026 the plans that list the CLI start with Plus at $20 a month, and Business seats cost $20 per user per month on a yearly contract.

For a founder buying for a small engineering team, the useful question is which mix of seats and keys matches how people actually work. Here is how the options compare, using OpenAI's Codex pricing page as checked on November 3, 2026.

The options side by side

Business also comes with no training on business data by default, which matters if your code is the product.

Seats buy an allowance, not a meter

On Plus, Pro and Business, Codex use counts against an allowance rather than a per-token bill. OpenAI says how far it stretches depends on the model, the size of each task and whether the work runs locally or in the cloud. It publishes rough local message ranges per five hours for Plus and standard Business seats, calls them estimates, and adds that weekly limits may apply. Pro plans have no five-hour limit right now.

That makes a seat easy to budget and hard to size. It costs the same in a quiet month as in a launch week, yet two engineers on the same plan can run out at very different speeds.

When someone hits the limit, the turn already running is allowed to finish. Plus and Pro users can then buy extra credits, Business workspaces on flexible pricing can buy workspace credits, and anyone can keep working locally on an API key at standard rates.

API keys for automation

The key route is metered. Every token is billed through your OpenAI Platform account, and OpenAI describes it as the fit for automation in shared environments like CI. Two trade-offs matter for a team:

• Cloud features such as GitHub code review and the Slack integration need ChatGPT sign-in, so they do not work on a key.
• OpenAI warns against exposing Codex in untrusted or public environments, so keep CI runners private.

OpenAI also says API token prices should not be used to estimate how much a subscription includes. Treat them as two separate budgets.

A simple buying order

1. Start each engineer on Plus and let them do normal work for two weeks.
2. Note who runs into limits, and how often.
3. Move only those people to Pro, or move everyone to Business once you need SSO and admin controls.
4. Put CI and scheduled jobs on an API key with its own spend limit.

Measure before you buy seats

Step 2 is where most teams guess. OpenAI's usage dashboard shows each person's current limits and /status shows the session inside the CLI, but neither keeps a running log you can compare across people and weeks.

The tracker we recommend for that is UsageAtlas, a free, open-source desktop app for Windows, macOS and Linux. Each developer runs it on their own machine, where it reads the signed-in Codex CLI and records session and weekly windows, reset times, credits where reported, tokens by model and an estimated cost over time. No account is needed for local tracking, prompts and responses are never read, and an optional free account saves the history. Code stays where it already lives.

Its cost figures are estimates, not an invoice, so OpenAI's billing remains the source for what you paid.

When to skip this

If you already run ChatGPT Enterprise, its analytics dashboard and Compliance API give admins organization-wide usage without asking anyone to install a desktop app.