Considered as ‘largest and most experienced’ model of developed by the company so far, a new proposal presents computational efficiency ten times better than the previous generation
Last Thursday (27), the He revealed that he is working on a new model of general artificial intelligence, called GPT-4.5. This model adopts a technique known as a mixture of experts, which is similar to the approach used by DeepSeek company. This innovation resulted in a significant 37% reduction in the so-called “hallucinations”, which are the moments when Chatbot creates false information compared to its predecessor, the GPT-4.
Currently, the GPT-4.5 is in the testing phase and is accessible to programmers and users who sign the plan Pro. Openai has not yet confirmed if the new technology was influenced by DeepSek. Although the model is more robust and complex, it has no advanced reasoning, which may limit its application to programming tasks.
In addition to the development of the GPT-4.5, OpenAi is looking to raise up to $ 40 billion in investments, which could raise its market evaluation to $ 300 billion. However, the company faces high operating costs, which leads it to consider the possibility of restricting access to the new model for developers.
Currently, access to GPT-4.5 is paid, but OpenAi does not rule out the possibility of revoking this access in the future, depending on the economic viability and the use of the model. This decision can impact the way developers interact with the new technology, reflecting the company’s concerns about financial sustainability.
“GPT-4.5 is a very large and intensive model in computing, making it more expensive and not a substitute for the GPT-4O [seu antecessor]. Because of this, we are evaluating whether we will continue to offer it in the API in the long run while we balanced current capabilities with the construction of future models, ”the company said.
*Report produced with the aid of AI
Posted by Victor Oliveira
