OpenAI fragments its stack

OpenAI is slicing ChatGPT into more layers, with different models, limits, prices, and even ads depending on which plan you use. (help.openai.com) On April 9, OpenAI said GPT-5.3 Instant Mini replaced GPT-5 Instant Mini as the fallback model users get after hitting GPT-5.3 Instant rate limits. The company said the fallback will not appear in the model picker. (help.openai.com) A separate help page says GPT-5.3 is now the default system for logged-in users, with “Instant” auto-routing some harder requests to GPT-5.4 Thinking. OpenAI retired GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking from ChatGPT on February 13, 2026. (help.openai.com) On April 16, OpenAI said it was beginning to roll out ads to Free and Go users in Australia, New Zealand, and Canada. The same release note said Plus, Pro, Business, Enterprise, and Education plans do not have ads. (help.openai.com) The coding side is being split up too. OpenAI updated Codex pricing on April 2 to charge by token use instead of by message for new and existing Plus, Pro, and Business customers, plus new Enterprise customers. (help.openai.com) That change turns coding access into a meter. OpenAI said Codex usage is now priced in credits per million input, cached input, and output tokens, while some existing Enterprise customers remain on the older per-message system until migration. (help.openai.com, help.openai.com) OpenAI also changed its seat structure for workplace plans. Business and Enterprise now include Codex-only seats with usage-based billing and no ChatGPT access, alongside standard seats that bundle ChatGPT and Codex together. (help.openai.com, help.openai.com) The company is also starting to ship models aimed at one field instead of everyone. On April 16, OpenAI introduced GPT-Rosalind, a life-sciences model for biology, drug discovery, and translational medicine, and said it is available as a research preview in ChatGPT, Codex, and the application programming interface for qualified customers. (openai.com) OpenAI said GPT-Rosalind is designed for evidence synthesis, hypothesis generation, experimental planning, and work across chemistry, protein engineering, and genomics. It also launched a Life Sciences plugin for Codex that connects to more than 50 scientific tools and data sources. (openai.com) The result is a product stack that now separates default chat, fallback chat, deep reasoning, coding, workplace seats, and domain models into different entitlements. OpenAI’s own help pages now describe a service where access depends less on one subscription and more on which model, seat, credit pool, and region a user is in. (help.openai.com, help.openai.com, help.openai.com)