3 Big Changes in GPT-5 Update That Users Didn’t Expect

Key Points

• 3 Big Changes in GPT-5 Update That Users Didn’t Expect
• GPT-5 introduces new user-selectable modes: Auto, Fast, and Thinking
• OpenAI restores GPT-4o and legacy models after community pushback
• GPT-5’s router fails to deliver the promised simplicity
• OpenAI now prioritizing custom personality and user preferences

When OpenAI officially released GPT-5 last week, expectations were sky-high. The model was marketed as a powerful all-in-one upgrade that would remove the hassle of choosing between multiple AI models.

The key feature was a model router, a smart internal tool meant to automatically select the best engine for each prompt.

The idea? Users shouldn’t have to scroll through a long list of models. The router would think, so you could focus on your work. But shortly after launch, it became clear that this vision didn’t match reality.

Updates to ChatGPT:

You can now choose between “Auto”, “Fast”, and “Thinking” for GPT-5. Most users will want Auto, but the additional control will be useful for some people.

Rate limits are now 3,000 messages/week with GPT-5 Thinking, and then extra capacity on GPT-5 Thinking…

— Sam Altman (@sama) August 13, 2025

OpenAI CEO Sam Altman took to X this week to announce new user-selectable modes within GPT-5:

• Auto – Automatically routes prompts, just as originally intended.

• Fast – Prioritizes speed over complexity, ideal for short answers.

• Thinking – Focuses on deeper analysis and reasoning, even if slower.

Instead of simplifying things, these new modes added more options, bringing back the complexity the update aimed to eliminate. In other words, the model picker is here to stay, and it’s more crowded than before.

Many users pointed out that the GPT-5 update felt more confusing, not less. The rollout faced numerous issues, prompting OpenAI to publicly respond just days after launch, admitting that the router was underwhelming and promising further refinements.

Source: OpenAI

Users Drive a Shift Toward Personality and Customization

One of the most surprising outcomes of the GPT-5 update has been the return of older models. GPT-4o Plus users in particular had strongly voiced frustration after their favorite models were suddenly removed. These models weren’t just tools—they had distinct personalities, which users had grown attached to over time.

As a result, OpenAI brought back legacy models like GPT-4.1, o3, and GPT-4o. GPT-4o is now the default model in the picker again, and users can manually enable others via settings. This was a major reversal and shows OpenAI is now taking user sentiment and emotional attachment seriously.

CEO Sam Altman admitted in a post that GPT-5’s current personality felt “too flat” and said a warmer version is in the works, one that won’t be as “annoying” as GPT-4o but still more engaging than GPT-5’s initial tone.

He also hinted at a bigger shift: “We just need to get to a world with more per-user customization of model personality.”

The rise of personalized AI experiences is already shaping the broader AI space. Anthropic recently saw users hold a literal funeral for Claude 3.5 Sonnet when it was taken offline, something previously unthinkable in tech circles.

This shows AI is no longer just about speed or intelligence. Users connect with how a model sounds, feels, and interacts—and removing that connection creates backlash.

OpenAI’s VP of ChatGPT, Nick Turley, acknowledged that the rollout wasn’t perfect but emphasized how fast the team is iterating. “We’re not always going to get everything on try #1,” he said on X, “but I am very proud of how quickly the team can iterate.”

We’re not always going to get everything on try #1 but I am very proud of how quickly the team can iterate. Keep the feedback coming! https://t.co/WoHqfWYfmG

— Nick Turley (@nickaturley) August 13, 2025

Even more interesting is how AI routing is becoming increasingly complex. It’s not just about fast vs slow anymore. Some users prefer verbose models. Others like a more minimalist tone. And then there’s the rising demand for emotionally intelligent AI.

OpenAI isn’t alone in rethinking user customization. Meta is also pushing innovation in AI experiences, particularly with Meta AI’s new voice waveform features, which enable more human-like and personalized audio interactions. The industry is moving in one direction: AI that adapts to the individual.

GPT-5 Update: Where It Stands Now

After a week of real-world feedback, the GPT-5 update has clearly evolved. While it introduced major improvements in speed and multi-modal reasoning, it failed in its mission to simplify the ChatGPT experience.

The automatic model router turned out to be less effective than hoped, and OpenAI is now giving users more control, not less.

The return of legacy models, the introduction of new response modes, and the focus on AI personality customization show that OpenAI is listening, though it wasn’t part of the original plan.

This moment marks a turning point in AI evolution: performance alone isn’t enough anymore. People want choice, familiarity, and connection with their AI. Whether that’s a fast response or a warm personality, OpenAI and others are being forced to design AI with human attachment in mind.

And if the last few weeks have shown anything, it’s that users will push back when their favorite models disappear, and companies will have to adapt, fast.