OpenAI has moved swiftly to contain discontent over last week’s release of GPT-5, restoring the earlier GPT-4o model as the default option for all paying ChatGPT customers and bringing back other legacy models such as GPT-4.1 and o3. In a series of posts on X, chief executive officer Sam Altman said the company would give “plenty of notice” before removing any popular model in future, acknowledging complaints that GPT-5 felt colder, produced inconsistent results and arrived without warning. The company is also tweaking GPT-5 itself. Users can now choose between “Auto”, “Fast” and a more resource-intensive “Thinking” mode that carries a 3,000-message weekly allowance and a 196,000-token context window. Altman promised a forthcoming update intended to give the new model a “warmer” personality while stopping short of claiming that GPT-5 meets artificial-general-intelligence thresholds, citing its inability to learn continuously. Beyond repairing trust, OpenAI is broadening ChatGPT’s reach. It has begun rolling out Connectors that let the chatbot pull information directly from Gmail, Google Calendar, Google Contacts, Dropbox, Microsoft Teams and other third-party services. The feature is available immediately to Pro subscribers, with Plus, Team, Enterprise and Edu tiers slated to follow. The company is also testing a lower-priced ChatGPT Go plan at 399 rupees (about €4) a month in India, supplementing its existing Plus tier at US$20. Together with the reinstated models and new integration tools, the moves underscore OpenAI’s bid to stabilise its core product and defend market share against rivals such as Anthropic’s Claude and Google’s Gemini after a rocky GPT-5 debut.