When OpenAI introduced GPT-5 last week, the company promoted it as a way to simplify the ChatGPT experience. The goal was for GPT-5 to serve as a “one size fits all” model, using a built-in router to automatically choose the best way to respond to questions. This unified setup was meant to eliminate the need for the model picker — a long, complex menu of AI options that CEO Sam Altman has openly criticized.
However, GPT-5 hasn’t fully delivered on that vision.
In a post on X Tuesday, Altman announced new “Auto,” “Fast,” and “Thinking” modes for GPT-5, all available through the model picker. Auto functions much like the original GPT-5 router, but OpenAI is also letting users bypass it, giving them direct access to both quicker and more deliberative AI models.
Altman Restores Access to Discontinued AI Models, Including GPT-4o, GPT-4.1, and o3
Along with GPT-5’s new modes, Altman announced that paid users can once again access several older AI models — including GPT-4o, GPT-4.1, and o3 — which had been discontinued just a week ago. GPT-4o now appears in the model picker by default, while the others can be enabled through ChatGPT’s settings.
In his post on X, Altman added, “We are working on an update to GPT-5’s personality that should feel warmer than the current one, but not as irritating (for most users) as GPT-4o. One thing we’ve learned over the past few days is that we need to move toward greater per-user customization of model personalities.”
GPT-5’s Model Router Falls Short of Expectations, Leaving Rollout Bumpier Than Anticipated
ChatGPT’s model picker remains as complex as ever, hinting that GPT-5’s model router hasn’t met user expectations as OpenAI had hoped. Anticipation for GPT-5 was immense, with many expecting a leap forward similar to GPT-4’s debut, but its rollout has proven bumpier than planned.
The removal of GPT-4o and other AI models triggered backlash from users who had grown attached to their unique responses and personalities — a reaction OpenAI hadn’t foreseen. Altman now says the company will provide ample advance notice if it ever decides to retire GPT-4o again.
GPT-5’s model router struggled at launch, leading some users to feel it underperformed compared to earlier OpenAI models. The issue pushed Altman to address concerns in a Reddit AMA, though the router still doesn’t seem to satisfy everyone.
On X, OpenAI’s VP of ChatGPT, Nick Turley, acknowledged the rocky start, noting, “We’re not always going to get everything on try #1, but I am very proud of how quickly the team can iterate.”
Routing prompts effectively is challenging, requiring the system to match a user’s preferences and query type, then decide within milliseconds which AI model should respond. Preferences go beyond speed — some users prefer the detail of one model, while others favor another’s contrarian style.
This human attachment to specific AI models is a relatively new phenomenon, not yet fully understood. In one case, hundreds in San Francisco held a mock funeral for Anthropic’s retired Claude 3.5 Sonnet, while in other instances, AI chatbots have been linked to vulnerable individuals spiraling into unhealthy mental states.
Overall, OpenAI still has significant work ahead in fine-tuning its models to meet individual user needs.
Read the original article on: TechCrunch
Read more: Launch of ChatGPT-5 Marks the ‘Start of a New Era for Humanity’