A pivotal post-launch interview
A week after GPT-5’s debut, OpenAI is rethinking how it retires models. In a new interview, Nick Turley—head of ChatGPT—explains why GPT-4o briefly disappeared, why it’s coming back, and how the company plans to manage model changes at unprecedented scale. He also describes features aimed at healthy usage, including overuse notifications, while noting ChatGPT reaches hundreds of millions of weekly users.
Turley told The Verge that OpenAI underestimated the emotional attachment many users felt for 4o. The attempt to simplify choices by replacing 4o outright with GPT-5 backfired for a sizable cohort of power users, leading the company to restore 4o availability (starting with certain tiers) and to work on a clearer deprecation schedule. The conversation offers rare insight into product strategy for one of the world’s most widely used consumer AI apps.
Why model changes are hard at this scale
OpenAI’s product philosophy favors simplicity: most users shouldn’t need to pick models; the system should select appropriately. But, as Turley noted, at “700 million” weekly users, even a minority preference translates into a large, vocal population. For professionals who tuned workflows, prompts, and expectations around 4o’s tone and behavior, abrupt removal disrupted work. The OpenAI GPT-5 rollout revealed that model personality isn’t a trivial detail—it’s part of the product contract.
Turley characterized the misstep as a learning moment. The team is now considering explicit deprecation timelines—akin to API lifecycles—and more configurability for power users, while maintaining a streamlined default for the majority. He also indicated ongoing work to bring some of 4o’s perceived “warmth” to GPT-5’s behavior, and highlighted a feature that lets users choose personalities as one way to address preference diversity.
Health, safety, and time well-spent
The interview also tackled concerns about overuse and sensitive topics. Turley said OpenAI consulted more than 90 experts across 30+ countries and has introduced “overuse notifications” that gently nudge users when their usage patterns look extreme. The goal, he said, is to build a product OpenAI could “unequivocally endorse to a struggling family member”—a striking standard for a consumer tech company, implying active efforts to mitigate unhealthy patterns, not just prevent overt harms.
At the same time, Turley reiterated that OpenAI’s business model does not depend on maximizing time-spent; the free-to-paid conversion path allows the team to prioritize user outcomes. That stance is notable amid broader debates about the societal impacts of AI chatbots and their role in advice-giving. The OpenAI GPT-5 rollout thus doubles as a design stance: shipping a more capable model isn’t enough; you must consider attachment, identity, and well-being.
What 4o’s return signals
Bringing 4o back signals a few things. First, model plurality inside a single product may be inevitable. Second, “feel” is a differentiator that can outweigh raw benchmark gains. Third, enterprises and pros need predictability; surprise model swaps can break workflows, governance reviews, and outputs that must be consistent over time.
For developers, the API has long had deprecation schedules. Extending a similar approach to end-user plans would mean calendar-based retirements, advanced notice, and perhaps a compatibility layer or “behavioral locks” that preserve a model’s tone for a defined period. That could resemble how cloud providers handle runtime or database version upgrades: configurable but time-bounded.
Competitive implications
The OpenAI GPT-5 rollout plays out against fierce competition. Rivals are pushing bigger context windows, memory features, and agentic capabilities. In this climate, product experience—latency, reliability, style, and trust—matters as much as peak capability. If OpenAI can combine GPT-5’s technical gains with user-controlled personality and clear deprecation guarantees, it maintains an edge with both mainstream users and power users who drive ecosystem mindshare.
What to watch next
Expect near-term “fast follows” refining GPT-5’s behavior, visibility into model selection, and clearer communications around model lifecycles. Watch for expanded personality controls and, possibly, enterprise-grade toggles that ensure consistent behavior during audits and long-running projects. Finally, monitor whether 4o remains available indefinitely or if OpenAI sets an eventual retirement date—with ample notice and migration tooling.
The interview marks a rare, candid window into the challenges of operating an AI product at planetary scale. OpenAI has learned that emotional attachment to AI is real, and that responsible product stewardship includes predictability, transparency, and safeguards for time well-spent. The next phase of the OpenAI GPT-5 rollout will test whether the company can harmonize pace of innovation with the stability users increasingly expect.
