OpenAI cancels the independent release of the o3 model and will launch GPT-5, which integrates multiple technologies, in the coming months

OpenAI's leader Sam Altman stated on Wednesday that the company is nearing the release of a new artificial intelligence model named GPT-4.5, with GPT-5 expected to be released in the coming months, indicating that the company has canceled the release plan for the o3 model.

Altman posted a series of messages on the social platform X, indicating that the model is internally referred to as "Orion" and is planned to be released "within weeks."

Analysts believe that the release of GPT-4.5 will mark the end of an era. Altman revealed that this will be the last "non-thinking chain model" (a model that does not rely on additional computational resources to simulate human reasoning) launched by OpenAI. Unlike the o3 and other reasoning models from OpenAI, non-thinking chain models tend to be less reliable in fields such as mathematics and physics. In recent years, OpenAI has adopted a reasoning-driven approach in some new models (such as o1 and o3) and will gradually move in this direction.

In his posts, Altman also stated that OpenAI plans to integrate the GPT series models (the core technology of ChatGPT) with the new o series models to create a system capable of using all tools, with the ability to judge when long-term thinking or quick responses are needed, and to competently handle a wider range of tasks. The goal is to simplify the user experience and avoid users having to make choices among complex model options.

Currently, if users want to force ChatGPT to use a specific model (such as o3-mini), they need to click the "Reason" button in the prompt bar or use the options in the model selector after the chatbot answers questions. If users subscribe to ChatGPT Plus or Pro, the dropdown menu may become lengthy, requiring selection among multiple models and intelligence levels.

He stated:

“We want AI to ‘run seamlessly,’ and we realize that the current model and product options have become overly complex. We hate the ‘model selector’ just as much as you do.”

Additionally, Altman announced that OpenAI is advancing GPT-5, with plans to release it in the coming months. This new version will integrate multiple AI technologies from the company, including o3. Therefore, OpenAI no longer plans to release o3 separately. Previously, OpenAI had released a smaller version of the system, o3-mini, but has not yet launched the full model.

Altman stated that OpenAI plans to offer free users unlimited chat access under the "standard intelligence level" after the official release of GPT-5, but still subject to "abuse thresholds" (Altman did not disclose specific details about the intelligence level and abuse thresholds). Meanwhile, ChatGPT Plus subscribers will be able to use the higher intelligence level of GPT-5, and ChatGPT Pro subscribers will be able to run an even higher intelligence level of GPT-5Altman said,

"The model will integrate functions such as voice, canvas, search, and deep research. One of our current core goals is to unify all models, creating an AI system that can integrate all tools, know when to think deeply, and be widely applicable to various tasks."

However, GPT-4.5 (Orion) encountered multiple performance challenges and technical issues during its development. Previously, several media outlets reported that the performance improvement of Orion was smaller compared to the advancement of GPT-4 over GPT-3, which means it does not have a significant breakthrough compared to GPT-4o