Spring cleaning in February: OpenAI removes the model 'zoo' from ChatGPT

Spring Cleaning in February: OpenAI Clears the "Zoo" of Models from ChatGPT

It seems OpenAI decided that democracy and wide choice are good, of course, but order comes first. If you're used to juggling different neural network versions in search of the perfect answer, I have news for you. The company declared February 13 a day of "general cleaning." And judging by the list of candidates for elimination, Sam Altman's broom is new and harsh.

What's Happening?

The situation resembles a restaurant menu that has grown so much that customers spend more time choosing a dish than eating it. OpenAI is removing an entire layer of models from ChatGPT's user interface. Under the knife are GPT-5 (in Instant and Thinking variations), as well as veterans of the scene — GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini.

Let's be honest: did you really understand the subtle differences between GPT-4.1 mini and o4-mini for everyday tasks? Probably not. Most users just clicked the "smartest" or "fastest" button. Now that choice will be made for you.

Why Now?

There are three reasons why this is happening, and none of them relate to the models being "tired."

First, inference costs. Maintaining a dozen different model weights in hot access for millions of users is burning money at industrial scale. Each model requires its own infrastructure, caching, and request logistics. Simplifying the lineup is a classic P&L optimization.

Second, choice paradox. UX research has been saying for years: if you give a user 10 options, they'll freeze. If you give one, but the best — they'll be happy. OpenAI is clearly moving toward a "single window" concept, where the system itself decides how much computing power to spend on your request without asking whether you want the Thinking or Instant version.

Third, data focus. To train the next generations (GPT-6?), companies need you to use current models. Your likes and dislikes on responses from outdated versions have much less value for RLHF (reinforcement learning from human feedback) than feedback on flagships.

What About Developers?

Here OpenAI showed rare prudence. Changes only affect "civilian" ChatGPT. Those building their applications through the API can breathe a sigh of relief. There will be no need to change code — at least for now. This makes sense: breaking production for corporate clients a day before Valentine's Day would have been too cruel even for a tech giant.

Usually, such a gap between UI and API versions signals that old models are being moved to "legacy" status. They will be available to machines, but hidden from human eyes so as not to spoil the product experience.

The Bottom Line:

The era of unlimited choice of chatbot versions is ending. OpenAI is clearly hinting: stop micromanaging models, just ask questions. Are we waiting for an announcement of a single universal model that will replace this entire zoo?