OpenAI tightens product lineup
OpenAI has been pruning which models appear inside ChatGPT while keeping API access steady, listing several retired models and limiting GPT-4o availability in some ChatGPT tiers. ( ). The company’s chief scientist described progress toward systems that can operate like a “research intern,” even as a Spanish report says OpenAI paused its planned Stargate UK data‑centre project over Britain’s high electricity costs and unsettled copyright rules. ( )
OpenAI is stripping older models out of ChatGPT while leaving the application programming interface largely unchanged. The shift became explicit in OpenAI help pages updated in April, which say GPT-5.3 is now the default chat system and several earlier models have been retired inside ChatGPT. (help.openai.com, help.openai.com) OpenAI says GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 Instant and Thinking were retired from ChatGPT on February 13, 2026. The same notices say those models remain available through the application programming interface, separating what ChatGPT users see from what developers can still buy. (help.openai.com, help.openai.com) The remaining exceptions were time-limited. OpenAI’s support pages say ChatGPT Business, Enterprise, and Edu customers kept GPT-4o inside Custom GPTs until April 3, 2026, after which GPT-4o was to be fully retired across all ChatGPT plans. (help.openai.com, help.openai.com) At the same time, OpenAI is narrowing the front-end menu while pitching a broader back-end system. Its GPT-5.3 help page says the model auto-switches across capabilities inside one “smart, fast single experience,” with GPT-5.4 Thinking positioned for harder reasoning tasks. (help.openai.com) That product cleanup is happening as OpenAI describes a more ambitious target for the software itself. Chief scientist Jakub Pachocki said the company is getting closer to systems that can work at the level of a “research intern,” according to reports published April 11. (timesofindia.indiatimes.com, newsbytesapp.com) The phrase points to software that can search, summarize, compare sources, and carry out multi-step desk work with limited supervision, rather than just answer one prompt at a time. OpenAI’s recent model notes make the same push in product language, describing systems built for “real-world work” across documents, spreadsheets, presentations, and software tools. (help.openai.com, timesofindia.indiatimes.com) The infrastructure needed for that vision is getting harder to line up outside the United States. Multiple outlets reported this week that OpenAI paused its planned Stargate United Kingdom data-center project, citing Britain’s high industrial power prices and unsettled copyright rules for artificial intelligence training. (cnbc.com, thenextweb.com, politico.eu) Politico reported that OpenAI had said it would explore leasing up to 8,000 advanced Nvidia chips from Nscale at United Kingdom sites in the first quarter of 2026. CNBC reported an OpenAI spokesperson said the company would keep exploring Stargate United Kingdom when regulation and energy costs support longer-term investment. (politico.eu, cnbc.com) The result is a company simplifying what ChatGPT users can pick, preserving continuity for developers, and tying its next step to bigger compute buildouts. OpenAI’s latest help pages and public comments point in the same direction: fewer visible model choices now, and more automated work from the surviving systems later. (help.openai.com, help.openai.com, timesofindia.indiatimes.com)
