OpenAI meters Codex in ChatGPT Business — usage now paid with admin-managed credits
- OpenAI has updated ChatGPT Business so Codex is no longer just a bundled perk — it now runs on workspace credits that admins manage. - The key shift is control: owners can buy credits, turn on auto-recharge, cap spend, and watch Codex usage analytics centrally. - That matters because OpenAI is turning team AI access into governed infrastructure, with model choice and usage limits varying by plan.
Coding help inside ChatGPT Business just got a lot more like cloud infrastructure. OpenAI has updated its Business workspace docs so Codex usage now sits behind a shared credit system, with admins buying credits, setting refill rules, and controlling how much the workspace can spend. That sounds like a billing tweak. But it’s really a product shift. Coding agents are being treated less like a flat subscription feature and more like a metered team resource. (help.openai.com) ### What changed in ChatGPT Business? The new Help Center flow is pretty explicit: if a Business workspace has Codex seats, expects extra Codex usage from standard seats, or wants workspace-level limits and analytics, it needs credits. Admins can add credits manually, enable auto top-up, and revi(help.openai.com) wrapped around it. (help.openai.com) ### Is Codex still included? Yes — but “included” now needs a footnote. OpenAI’s Codex plan page says Codex is available on Plus, Pro, Business, and Enterprise/Edu plans. But Business workspaces can mix seat types, and those seat types behave differently. A standard ChatGPT Business seat has a fi(help.openai.com)ng are no longer the same thing. (help.openai.com) ### Why do credits matter so much? Because credits let OpenAI separate permission from consumption. A team can give people access to Codex without pretending every user will consume the same amount of compute. If someone blows through per-seat limits, the workspace can keep them going from a shared credit pool — if t(help.openai.com)s they need to unblock over-limit usage or they’re running Codex seats. Basically, the subscription gets you in the door. Credits decide how far you can go. (help.openai.com) ### What’s happening under the hood on pricing? OpenAI also changed how Codex is priced. The Codex rate card says that on April 2, 2026, pricing moved from per-message billing to API-style token usage for Plus, Pro, Business, and new Enterprise plans. Then on April 23, 2026, that update was extended to(help.openai.com), and govern than a fuzzy “message” allowance. (help.openai.com) ### Why bring Enterprise and Education into this? Because the same docs now make the broader pattern clearer. OpenAI’s Enterprise and Edu model-limits page says model availability and limits vary by plan and by workspace settings. Business has its own models-and-limits page too. So the company is nudging organizations toward a familiar(help.openai.com)d budget you actually have. (help.openai.com) ### Is this just billing hygiene? Not really. This is governance. Admin-managed credits mean finance, IT, and workspace owners can all put guardrails around expensive AI behavior without banning the tool outright. Usage analytics make the coding assistant visible as an organizational resource, not just an emplo(help.openai.com)ing new Codex seats — shows OpenAI is actively trying to seed that usage model. (help.openai.com) ### What does this mean for teams? Teams now have to think about Codex the way they think about cloud compute or SaaS licenses. Who gets access? Which seat type makes sense? Do you want overage unblocked automatically? How much budget are you willing to expose? Those are admin questions, not just product questions. And once they exist, coding agents (help.openai.com)ged infrastructure. (help.openai.com) ### Bottom line OpenAI didn’t just add a payment toggle. It made Codex inside ChatGPT Business legible to administrators — budgeted, observable, and controllable. That’s a small interface change with a big implication: enterprise AI is moving from “everyone gets a bot” to “every workload gets a policy.” (([help.openai.com)siness))