Why OpenAI is retiring GPT-4o in ChatGPT despite protests

A model can disappear in a product update without anyone noticing. But not this time. The anger is not just about performance and refinement. It is about losing a certain personality and the creative flair that accompanied one’s daily work.

On January 29, the announcement came with a date: GPT-4o will retire in ChatGPT on February 13, 2026, along with GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and the previously announced retirement of GPT-5 (Instant and Thinking).

The message is that this is a ChatGPT decision, not an API shutdown. For now, the models are still available through the API.

The official justification: What is being changed 

The retirement of GPT-4o is framed as the end of an overlap period. GPT-4o was reintroduced after the GPT-5 rollout because some paid users wanted more time and enjoyed working with its specific style. Now, GPT-5.1 and GPT-5.2 are presented as the replacements, offering greater personality and customization.

The key statistic is that only 0.1% of daily users still choose GPT-4o, and most traffic has shifted to GPT-5.2. Even if that is a small percentage, at this scale, it represents a lot of user engagement. The retirement is happening now because the upgrades are ready and the focus is shifting to what most people use.

The backlash: Where the anger comes from 

The anger is not coming from users who casually click the default. It is coming from users who intentionally choose the models for their workflow.

The most common complaint is that GPT-4o felt more natural to interact with. The announcement even admits that some users preferred GPT-4o’s conversational style and warmth for creative ideation, a preference that influenced the newer models. This confirms that the reaction is not imagined.

But there is a second thread that is easy to miss: Some people aren’t arguing that GPT-4o is better overall. They are arguing that choice is being reduced. This is fueled by media coverage highlighting calls for cancellation, treating this as a test for whether the company listens to power users after the last reversal.

That is why there are two sides to this debate.

The structural incentives for shutting down GPT-4o 

The product decision makes sense, but there is also a structural reason for this.

First, model sprawl is expensive. Maintaining multiple models in a consumer product isn’t trivial. It is not just a UI decision; it involves routing, monitoring, and safety tuning for multiple stacks. When only one model sees significant usage, the rest become harder to justify. That 0.1% number suggests GPT-4o is now seen as overhead.

Second, there is a business incentive for standardization. A single experience is easier to improve, communicate, and keep consistent. That matters when you are relying on infrastructure from giants like Microsoft and Nvidia. The economics reward consolidation.

Third, the move signals a desire to reduce behaviors that frustrate mainstream users, such as unnecessary refusals or overly cautious responses. That kind of tuning is easier when fewer models define the core experience.

What this means for power users 

If you use GPT-4o for tone or flow in creative work, the plan before February 13 is simple. Save your patterns. Save your prompts. Save how you request feedback. Those habits transfer better than any single model tag.

The bet is that the majority will accept the compromise: less choice, more consistency, and a faster route to improving a single model. The backlash shows that for some, a model is not just a tool, it is a relationship with a particular voice.