Following a rocky rollout of GPT-5 and the removal of the option to select previous models, OpenAI faced a storm of protests. Now, the company is quickly backtracking, bringing back older models, introducing new control options, and admitting that it needs to further customize the AI’s personality.
Article contents:
- Introduction: A launch that didn’t go as planned
- The “one model to rule them all” plan and its backlash
- User protests and OpenAI’s quick response
- New control options: “Auto”, “Fast”, and “Thinking”
- Return of old models: The “Show additional models” toggle
- Admission of failure: “We won’t always get it right on the first try”
- Working on a “warmer” personality for GPT-5
- Conclusion: A lesson in listening to users
- Frequently asked questions
Introduction: A launch that didn’t go as planned
OpenAI, a leader in the field of artificial intelligence, experienced a rapid retreat and strenuous attempts to contain user anger following what was described as an “ill-considered” launch of its new GPT-5 model. The plan was ambitious: to unify the experience under the umbrella of a single intelligent model, but reality proved that power users still prefer to control their tools.
The “one model to rule them all” plan and its backlash
OpenAI’s vision for GPT-5 was to be a single model to rule them all, intelligently routing user prompts to the appropriate destinations without requiring user intervention. In other words, the company wanted to remove the model selection dropdown (such as GPT-4o, GPT-4.1, etc.) and let GPT-5 determine the best way to handle each request. However, this move, meant to simplify the experience, backfired, with many users—especially developers and advanced users—feeling they had lost control over a tool they rely on heavily.
User protests and OpenAI’s quick response
It wasn’t long after the launch of GPT-5 last week that social media and developer forums filled with complaints. Users protested the removal of previous models, pointing out that each model has strengths and weaknesses, and that they prefer to choose the most suitable model for their specific task. After a weekend spent dealing with these protests, CEO Sam Altman ultimately announced that the 4o model “is back.” But the company didn’t stop there, pushing forward with efforts to respond to user feedback, even if it meant moving away from its original goal of a unified GPT-5 model.
New control options: “Auto”, “Fast”, and “Thinking”
In an attempt to appease users demanding more control, Altman announced the addition of new options for GPT-5. He wrote this morning: “You can now choose between ‘Auto’, ‘Fast’, and ‘Thinking’ for GPT-5.” He added: “Most users will want to use ‘Auto’, but the extra control will be useful for some people.”
These new options are:
- Auto: This is the default mode that reflects OpenAI’s original vision, where GPT-5 decides how to handle the request.
- Fast: This mode likely prioritizes response speed and may use smaller or less complex models.
- Thinking: This mode is dedicated to complex tasks that require deeper “thinking.” It comes with a higher usage limit of 3,000 messages per week.
Return of old models: The “Show additional models” toggle
Most importantly, OpenAI responded to the key user demand. Altman confirmed that “4o is back in the model picker for all paid users by default,” and promised to give “ample notice” if the company decides to retire it in the future. To satisfy more demanding users, the company added a new option in the settings. Altman said: “Paid users now have a ‘Show additional models’ toggle in their web ChatGPT settings, which will add models like o3, 4.1, and GPT-5 Thinking mini.”
Admission of failure: “We won’t always get it right on the first try”
These updates are an implicit admission that GPT-5’s smart routing system did not live up to all users’ expectations, and that the company needs to put in more effort. Following Altman’s announcement, Nick Turley, VP and Head of ChatGPT, admitted: “We won’t always get everything right on the first try,” but added that he is “super proud of how fast the team has been iterating and improving.”
Despite these issues, adoption of GPT-5 has not stopped. Microsoft confirmed yesterday that the model is now rolling out in GitHub Copilot for Visual Studio, describing it as “OpenAI’s most advanced model yet.”
Working on a “warmer” personality for GPT-5
Concluding his remarks, Altman revealed another aspect of user feedback the company is working on. He said that OpenAI is “working on a personality update for GPT-5 that should feel a bit warmer than the current one, but not annoying (to most users) like GPT-4o.” He added: “One lesson from the last few days is that we really need to get to a world where model personality is much more customizable per user.”
Conclusion: A lesson in listening to users
This situation represents an important lesson for OpenAI and other tech companies: in the race toward innovation, the preferences and needs of the core user base should not be ignored. The quick retreat and response to user feedback demonstrate maturity on OpenAI’s part, but the rocky launch of GPT-5 shows that the road to ideal artificial intelligence is still long, and that human control remains an essential part of the equation.
Frequently asked questions
Q1: What was the main issue with the GPT-5 launch?
A: OpenAI removed the option to select older models (like GPT-4o), forcing users to rely on GPT-5 to route their requests, which displeased many users who preferred manual control.
Q2: How did OpenAI respond?
A: OpenAI brought back the GPT-4o model as a default option and added a toggle to show more older models. It also introduced three operating modes for GPT-5: Auto, Fast, and Thinking.
Q3: Is OpenAI working on improving the GPT-5 personality?
A: Yes, Sam Altman said they are working on an update to make the model’s personality “warmer” and less formal, while acknowledging the need for greater customization for each user in the future.
