OpenAI widens GPT-5 access

OpenAI is widening access to GPT-5 inside ChatGPT, but it is not flipping one giant switch. Its help pages now say GPT-5 is “slowly rolling out” across Plus, Pro, Team, and Free plans worldwide on web, mobile, and desktop, which means users can see the same model name spread plan by plan instead of arriving everywhere at once. (help.openai.com) That staggered rollout tells you two things at once. OpenAI wants GPT-5 to become the default experience for ordinary ChatGPT users, but it still wants room to meter traffic, watch reliability, and tune limits by tier while the rollout is live. (help.openai.com) The enterprise side is moving more carefully than the consumer side. OpenAI’s release notes say some model access and app controls in ChatGPT Enterprise and ChatGPT Edu are default-off and must be enabled by workspace admins, which keeps new features from automatically reaching every employee in a company or school. (help.openai.com) Those app controls matter because “connectors” are now called “apps,” and those apps let ChatGPT pull data from outside services like Box, Notion, Dropbox, or SharePoint. OpenAI says app availability can vary by plan and region, and enterprise admins can manage who gets access through workspace settings and role-based permissions. (help.openai.com 1) (help.openai.com 2) OpenAI is also drawing a sharper line around training data. Its apps documentation says data from connectors is not used to train models by default for ChatGPT Business, Enterprise, and Edu customers, while Free, Plus, Go, and Pro users may allow model training through the “Improve the model for everyone” setting. (help.openai.com) The second half of the story is billing. On April 2, 2026, OpenAI updated its Codex rate card so ChatGPT Business and new ChatGPT Enterprise plans moved from per-message pricing toward rates aligned with application programming interface token usage, which means customers are charged more like a cloud meter than an all-you-can-send chat box. (help.openai.com) A token is the small unit OpenAI uses to count text, like charging by ingredients instead of by plate. When pricing shifts from messages to tokens, a short request and a giant code-heavy request no longer cost the same just because each one was “one message.” (help.openai.com) OpenAI paired that pricing change with a new seat structure. Its flexible-pricing documentation says ChatGPT Business and ChatGPT Enterprise now have two seat types as of April 2, 2026: a standard ChatGPT seat and a Codex-only seat, so a company can give some workers full chat access and others only the coding agent. (help.openai.com) The Business billing page adds another concrete shift: standard ChatGPT seats got a USD $5 per month price cut on April 2, 2026, while Codex seats are billed on usage and require workspace credits. OpenAI also says eligible Business workspaces can earn up to $100 in Codex credits per newly added Codex seat, capped at $500 per workspace. (help.openai.com 1) (help.openai.com 2) That is the bigger pattern behind the rollout. OpenAI is making the flagship model broader at the front door while making advanced agent work more granular at the cash register, so ChatGPT starts to look less like one subscription and more like a bundle of metered compute, controlled integrations, and role-based access inside a workplace. (help.openai.com 1) (help.openai.com 2)