OpenAI adds business credit tiers

OpenAI just made a quiet but important change to ChatGPT Business. Companies on flexible pricing can now buy credits to push past the built-in limits for local tasks, instead of simply hitting a wall and stopping. That sounds like a billing tweak. It is — but it’s also a product-design signal. OpenAI is making access limits, pooled credits, and spend controls part of the normal operating model for business AI, not an edge case. (help.openai.com) ### What actually changed? The new piece is simple on the surface: ChatGPT Business workspaces using flexible pricing can purchase credits to increase access above the included limits for local tasks. OpenAI folded this into the Business release notes, which means this is no longer buried in enterprise sales conversations — it’s a documented feature for self-serve business customers. (help.openai.com) than it sounds? Because ChatGPT Business is no longer just “pay per seat and use the app.” OpenAI’s help docs now describe two seat types for Business and Enterprise — a standard ChatGPT seat and a Codex-only seat introduced on April 2, 2026. Credits sit on top of that system and unlock extra access to advanced features like Deep Research, Thinking models, Image Gen, Advanced Voice, an(help.openai.com 1) (help.openai.com 2) ### How do the credits work in practice? For Business, users still get per-seat usage limits. If someone burns through that included allowance and the workspace has purchased credits, usage can continue from a shared pool. If no credits are available, the feature gets blocked and the user can ask an admin to add more. Admins can also turn on automatic recharge, set a minimum balance, ref(help.openai.com)ms a lightweight internal billing system inside ChatGPT itself. (help.openai.com) ### Why does this matter for developers too? Because the same company is telling developers, very explicitly, that limits are real infrastructure constraints. OpenAI’s API docs say rate limits are enforced across RPM, RPD, TPM, TPD, and IPM, and long-context requests can have separate limits. Those limits exist at the organization and project level, and some model families share pools. That means “just retry on 429” was never a complete strategy. (developers.openai.com) ### So what breaks when teams ignore that? At small scale, not much. At larger scale, a lot. The practical guide circulating this week makes the point clearly: retries can burn more quota, synchronized workers can slam the same ceiling again, and one noisy internal consumer can starve everyone else when limits are shared. The catch is that usage controls are now showing up in both the API world and the ChatGPT workspace world. (developers.openai.com)s, usage caps, admin alerts — before customers hit the ceiling. (getmaxim.ai) ### What about Codex? Codex is part of the story because credits are now explicitly tied to it. OpenAI says Codex seats require credits for activity, and workspace analytics now include Codex usage. So this is not only about chat volume. It’s also about how companies budget and govern agentic coding work inside the same workspace. (help.openai.com)h — but the product angle is the more interesting one. OpenAI is exposing a truth that a lot of AI vendors prefer to hide: capacity is not infinite, premium features are not flat-rate in any durable sense, and business adoption needs controls around who can spend what, where, and how fast. Credits are the monetization layer. The real shift is that graceful degradation is becoming part of the user experience. (help.openai.com) ### Bottom line This update turns “limits” from a background technicality into a visible business setting. If you’re rolling out ChatGPT Business now, onboarding, budgeting, and fallback behavior all need to be designed together — because OpenAI clearly is. (help.openai.com)