ChatGPT’s Model Picker Returns With New Complexity
▼ Summary
– OpenAI launched GPT-5 with the goal of simplifying ChatGPT by using a “one size fits all” model router, eliminating the need for a complex model picker.
– GPT-5 did not fully unify AI options as planned, introducing new settings (“Auto,” “Fast,” and “Thinking”) and reinstating deprecated legacy models like GPT-4o.
– Users expressed backlash over the removal of older models, leading OpenAI to promise advance notice for future deprecations and explore per-user customization of AI personalities.
– GPT-5’s model router faced technical issues at launch, causing performance concerns and prompting OpenAI to acknowledge the need for iterative improvements.
– Human attachment to specific AI models, like Claude 3.5 Sonnet, highlights the challenge of aligning AI preferences with diverse user needs and behaviors.
OpenAI’s latest GPT-5 update brings back the model picker with new customization options, despite initial promises of a streamlined experience. The company had positioned GPT-5 as an all-in-one solution that would automatically route queries to the most suitable AI model, eliminating the need for manual selection. However, user feedback and technical challenges have forced a rethink, reintroducing complexity through multiple response modes and legacy model access.
Sam Altman recently announced three new settings for GPT-5: “Auto,” “Fast,” and “Thinking.” While Auto mirrors the original vision of an intelligent router, Fast prioritizes quick responses, and Thinking engages deeper processing for complex queries. Paid subscribers can also reactivate older models like GPT-4o and GPT-4.1, which were unexpectedly removed last week, a decision that sparked significant backlash from users attached to their quirks and personalities.
The reintroduction of these choices suggests GPT-5’s routing system hasn’t fully met expectations. Early performance issues compounded frustrations, with some users reporting slower or less accurate responses compared to previous models. Altman acknowledged these growing pains, emphasizing ongoing improvements to balance warmth and efficiency in GPT-5’s tone while exploring personalized customization.
Technical hurdles remain in prompt routing, a process requiring split-second decisions about which model version best suits a query. Speed isn’t the only factor, user preferences vary widely, from verbose explanations to unconventional perspectives. This emotional connection to AI behavior has become surprisingly strong, evidenced by events like the public mourning for Anthropic’s retired Claude 3.5 Sonnet model.
OpenAI’s struggle highlights a broader challenge in AI development: balancing simplicity with flexibility. While unified models promise convenience, diverse user needs and unpredictable attachments complicate standardization. The company now plans longer deprecation notice periods and deeper personalization features, recognizing that one-size-fits-all solutions rarely satisfy everyone in this rapidly evolving space.
(Source: TechCrunch)
