ChatGPT Business and Enterprise: Codex and Programmatic Use

ChatGPT Business and Enterprise do not include OpenAI API credits; the API remains a separate, per-token product. What they do include is Codex on every seat, and Codex runs programmatically through codex exec, its documented non-interactive mode. For a team, that means the flat seats you already pay for can back real automated workloads, with workspace admin controls on top.

Here is what the business plans actually include, the seat capacity math, and the account rules that matter when a team automates.

What Business and Enterprise actually include

ChatGPT Business (the plan formerly sold as Team) lists at $25 per seat per month on annual billing, $30 month to month; Enterprise is custom-priced through sales. Check OpenAI’s pricing page for current figures. The relevant inclusions:

Capability	Business	Enterprise
Flagship chat models	Yes	Yes
Codex (CLI, IDE, cloud)	Yes, per seat	Yes, per seat
Workspace admin controls	Yes	Yes, expanded
SSO	Yes	Yes
Excluded from training by default	Yes	Yes
OpenAI API credits	No	No (separate contract)

The last row is the one that surprises buyers. Upgrading the workspace does not touch your API bill, because they are different products with different meters; the split is the same one we walk through in why your OpenAI API bill is separate from ChatGPT and what ChatGPT Plus actually includes, just multiplied by seats.

Codex is the programmatic surface

Buying ChatGPT Business does not buy API credits; it buys Codex on every seat, and Codex is the programmatic surface.

Three documented facts make that useful. OpenAI’s README recommends ChatGPT sign-in for the Codex CLI as part of Plus, Pro, Business, Edu, and Enterprise plans (github.com/openai/codex). codex exec is the CLI’s documented non-interactive mode for scripts and CI (developers.openai.com/codex/noninteractive). And device-code sign-in for headless machines is documented too (developers.openai.com/codex/auth). Intended functionality, with the standing caveat that applies to everything in this category: OpenAI has the final call on how its services may be used.

Each seat’s Codex usage is governed by rolling windows whose exact numbers vary by plan and model and drift over time; we deliberately do not reprint them. The mechanics are in our Codex usage limits explainer, and current figures live on OpenAI’s pricing page.

Seat capacity math for automated workloads

Ten seats are ten independent windows, not one shared pool. That shape matters: capacity scales sideways with seats, and one person’s heavy week does not drain a teammate’s window.

A planning sketch, with the assumption stated. Our estimates put an individual Plus window around $700 of API-equivalent work per month. If a Business seat’s Codex window behaves in that family:

10 seats × $25            = $250/mo flat
10 windows × ~$700 equiv. ≈ $7,000/mo API-equivalent capacity (estimate)

Treat every number on the right as an estimate, not a guarantee: Business windows are not identical to Plus, OpenAI adjusts limits, and a seat whose human codes with Codex all day has less window left for automation. The planning posture that works: automate against specific seats, watch observed consumption for a week, then size. Per-seat request logs make that observation trivial.

The account rules teams actually need

OpenAI’s terms prohibit sharing an account or making it available to anyone else, and that rule does not soften inside a workspace. The patterns, plainly:

• Fine: each engineer runs Codex under their own seat, interactively or in scripts.
• Fine in shape: a dedicated account the company owns, used by one automated workload, one sign-in, one container. One workload tied to one account your org owns is a different thing from five engineers passing around one login.
• Not fine: shared credentials, pooled logins, or reselling seat capacity outside the org.

The full clause-by-clause reading is in sharing an OpenAI account: what the terms say. Admins keep the levers either way: workspace controls govern who has Codex access, SSO governs who can sign in at all, and disconnecting a seat kills its sessions.

Where ProxyLLM fits for teams

Codex Hosted gives each connected account its own isolated container, signed in through OpenAI’s device-code flow; we never see the password and never pool accounts across customers or seats. Connect the seats you automate, set the fallback order (second account, then your own API key), and the request log shows which lane served every call. The $129 flat fee covers the workspace, not per seat, with no inference markup.

If your team’s API bill is the reason you are reading this, the calculator maps it against seat-backed capacity in thirty seconds.