Openai, a combination of new AI model releases of GPT-4.5

Unlock the editor’s digestion free

FT editor Roula Khalaf, chooses his favorite stories in this weekly newsletter.

Openai, the largest artificial model of artificial intelligence, with a fewer “hallucinations”, was beginning with the latest in a flutter of AI by US technological groups.

On Thursday, San Francisco Sağışık, GPT-4.5, presented a long-awaited update to Chatgpti of his popular product. In early tests, its hallucination rate, where AI The systems were inaccurate information, 37 percent compared to about 60 percent in the predecessor GPT-4O.

With GPT-4.5, Open Despite the formation of high, expensive small products as R1, the Chinese start-up DeepSeek continues to bet on large, cheaper and easier for developers.

Tech groups come to grow as competition in the recent weeks to start running their latest models. Anthropically revealed Clod 3.7 Sonnet on Monday, which followed the last week of the last week of Elon Musk’s Xai.

Openai on Thursday, Thursday, GPT-4.5 “leads to a more profound knowledge and a deeper understanding, reduced hallucinations and more reliable on large matters.”

“The company comes from novel opportunities, each new rule of the compartment scale,” the company said.

Openai, leading the AI ​​industry, was at a global competition to increase the first billions to finance large models of ten billions of ten billions of investors.

In talks with SoftBank and other investors, to increase $ 40 billion in an assessment $ 300 billionincluding new money. Anthropic, as well as $ 3.5 billion in an estimate of $ 60 billion, and said two knowledge of this process.

However, the great costs of the development of larger models allowed OpenaI to think of GPT-4.5 to join the GPT-4.5 among the price of opponents.

Openai said that GPT-4.5 will be accessible to the developers who pay to pay to use Openai models through a preview, application programming interface (API).

AI groups generally earn income through paid API access and individual subscribers. Openai said the developers would see how the powerful model uses and to offer them, taking into account the high price.

The company said: “GPT-4.5 is a very large and calculated model, a substitute for GPT-4O (predecessor). Therefore, we appreciate the API continues to be a long-term balance that we support current capabilities.”

Openai CEO Altman said that it costs more than $ 100 million for GPT-4 trains and requires more calculation power to scale and operate such expenditures and the size and capacity of models.

After the announcement, in an article in the X, Altman said that the company’s “GPU came out of GPU”, and the chips required to train AI systems and train.

“It’s not how we want to work, but it is difficult to predict the growth rate that causes GPU shortcomings,” he said, “he expected to receive more in the coming weeks.

He also stressed that he is not aimed at justifying GPT-4.5 and does not beat the industrial criteria, but “there is a magic in a different intellect and before.”

Additional report by George Hamond in San Francisco and Melissa Heikil in London