Too many models, too much confusion: Openai promises to simplify your product line

Openai plans to “simplify” its model offers, even if it is preparing to send its last non-limitation model, GPT-4.5. This is a rare entry from a technology company that does not differentiate its product publications for customers.

Openai CEO Sam Altman to X To give product -roadmap updates and to recognize that some of the company’s latest publications have confused the waters.

“We would like to better share our intended roadmap and a much better job to simplify our product offers,” Altman wrote. “We want AI to only work for her. We recognize how complicated our model and product offers have become. “

Altman took a special roof against Chatgpts Model Picker, in which users tap a drop-down selection to select from published models. The number of models that a user can choose depends on its subscription level.

Altman said that his team “hates the model picker as well” and will bring “Magic Unified Intelligence” back, which is probably referring to a single model that applies a platform. Altman said that the company plans to integrate its technology, including the O series of models, into GPT-5 when it comes out on Chatgpt and its APIs.

Last year openai GPT-4O publishedPresent O1Present O3 and O3-Mini Models. New Chatgpt subscriptions were also started, including the $ 200/month Chatgpt Pro.

In addition, the company has published operatorDeep Research, Chatgpt tasks and a variety of other new functions, which are sometimes only available via certain models or paid levels.

Waiting for the publication of Openas GPT-4.5 and 5

Altman said Openas Next step was to have published its last model without a chain (cot). It was called Orion, although he did not explain what new skills GPT-4.5 will have.

“After that, it is a top goal for us to combine models of the O series and models of the GPT series by creating systems that can use all of our tools, when they have to think for a long time or not and generally useful for one A very long range can be tasks, ”said Altman.

If the company publishes GPT-5, the shipping of the O3 argument is discontinued as an independent model.

Openai announces that some users may not need the full functions of their latest models, Altman admitted, which is why the company has created a model picker at all. GPT-5’s skills are tailored to the subscription level to replace the drop-down company, he said.

“The free chatt level receives unlimited chat access to GPT-5 in the standard intelligence setting (!!), which is subject to abuse swelling,” said Altman. “Subscribers can also run GPT-5 at a higher intelligence level, and Pro subscribers can perform GPT-5 at an even higher level of intelligence. These models are contained language, canvas, search, deep research and much more. “

Altman did not say when GPT-4.5 or GPT-5 could be released.

Customer confusion

Openaai’s decision to simplify how models are presented on a persistent problem: not everyone understands what every model should do and what it does. Openai is not alone. If I didn’t report on industry every day, I could not distinguish between the Gemini models and tell them why Claude’s Sonett is better than Haiku.

While this confusion is more common among consumers, it also makes it much more manageable for electricity users such as AI engineers.

They have to have a functioning encyclopedia, from which models can do what and which updated. If you integrate the latest functions into a model and only enable more complex skills based on use, many developers can be facilitated to experiment faster.