OpenAI sets enterprise model limits

OpenAI has started spelling out something enterprise buyers have been asking for all along — what exactly is unlimited, what is metered, and where the bill starts running. The change is not a splashy product launch. It is documentation. But the documentation matters because it turns ChatGPT Enterprise, Edu, and Codex from fuzzy bundles into products with clearer operating rules. Over the last few weeks, OpenAI updated its help center to publish explicit model limits for Enterprise and Edu, a formal ChatGPT rate card, and a token-based Codex rate card that now applies across most paid tiers. ### What actually changed? Two dates matter. On April 2, 2026, OpenAI says it introduced Codex-only seats for ChatGPT Business and Enterprise and switched Codex pricing to token-based credits instead of per-message pricing. On April 23, 2026, that new pricing was extended to all existing Enterprise plans, including Edu, Health, Gov, and Teachers. ### What did OpenAI publish? There are really three pieces. (help.openai.com) First, a models-and-limits page for ChatGPT Enterprise and Edu that explains which models are available, which ones are retired, which ones are disabled by default, and how “virtually unlimited” access works. Second, a ChatGPT rate card for flexible pricing on advanced features. Third, a Codex rate card that lays out credit costs by model and token type. Basically, OpenAI moved key commercial terms out of sales conversations and into public support docs. (help.openai.com) ### So is Enterprise still “unlimited”? Yes and no — and that is the whole story. The Enterprise plan still says it offers virtually unlimited GPT-5.5 messages, especially for Instant models. But the same documentation now makes clear that advanced features and some higher-end model usage can draw from shared workspace credits under flexible pricing. In other words, “unlimited” is real for a broad class of normal use, but not a blank check for every expensive workflow. (help.openai.com) ### How does the new Codex pricing work? Codex is now priced like an API product wearing a ChatGPT suit. OpenAI lists credits per million input, cached input, and output tokens. GPT-5.5 is priced at 125 credits for input tokens, 12.5 for cached input, and 750 for output. GPT-5.4, GPT-5.4-Mini, GPT-5.3-Codex, and GPT-5.2 each have their own rates, and fast mode can cost more. OpenAI also says average Codex spend runs about $100 to $200 per developer per month, with big variation by model and workflow. (help.openai.com) ### Why do Codex-only seats matter? Because they separate “AI for coding” from “full ChatGPT access.” Enterprise now has standard ChatGPT seats and Codex seats. Codex seats are usage-based, have no fixed monthly user fee, and only unlock Codex — not the broader ChatGPT workspace. That makes procurement more modular. A company can give a lot of developers coding-agent access without buying every one of them a full enterprise chatbot seat. (help.openai.com) ### What does this mean for admins? Admins get more levers, but also more budgeting work. Enterprise and Edu use shared credit pools bought at the contract level. Owners can set spend controls by group, get alerts when thresholds are hit, enable overages, or pause advanced features when the pool runs out. That is a very different posture from the old SaaS habit of buying seats and not thinking much about marginal usage. (help.openai.com) ### Why does this matter beyond OpenAI? Because this is what enterprise AI is turning into — governed consumption. The winning product is not just the smartest model. It is the model wrapped in permissions, routing rules, spend controls, and predictable billing. OpenAI’s docs make that shift visible. The model is still the headline. But the rate card is starting to run the business. (help.openai.com 1) (help.openai.com 2)