Openai introduced its latest GPT-4.5 language model, which, according to the company, is the most advanced and natural today, although the most expensive. The model has an expanded knowledge base, increased emotional intelligence and an improved ability to follow the user’s intentions, which makes it more effective in tasks, such as writing texts, programming and solving problems.
The co-founder and general director of OpenAI Sam Altman noted that although the GPT-4.5 is not a model for logical reasoning and is not intended to achieve high results in tests, it has a special “magic”, making the interaction more warm and intuitive. Altman emphasized that the model is able to conduct conversations, felt as communication with an attentive person, and expressed the joy of the possibility of users to test it.
However, Altman also pointed to the high cost of the GPT-4.5, calling it a “huge and expensive” model. Due to the lack of graphic processors (GPU), the model will initially be available only to ChatgPT Pro users, and the expansion of access to other subscriptions is planned in the near future.
GPT-4.5 was developed using scalable teaching methods, which made it possible to improve its ability to recognize templates and creative insights. The model also demonstrates a decrease in the frequency of “hallucinations” – cases when AI provides inaccurate information. Early testing showed that interaction with GPT-4.5 is felt more natural due to its expanded knowledge base and improved emotional perception.
Currently, GPT-4.5 is available as a research version for ChatGPT PRO users and developers with paid access to the API. Openai plans to gradually expand access to the model for other categories of users in the coming weeks. The company also expressed interest in observing how users will use GPT-4.5 in unexpected scenarios to better understand its capabilities and restrictions.