Openai discovers GPT-4.5 against the background of the new AI Models Models Models

Unlock Digest Editor free

Halef’s roller, FT editor chooses his favorite stories in this weekly newsletter.

Openai launched its largest artificial intelligence model with fewer “hallucinations”, the latter in a flurry of AI technological groups.

On Thursday, the company based in San Francisco introduced the GPT-4.5, its long-awaited technology update underlying the popular product. In the early trials, its speed of hallucinations where AI The systems create inaccurate information, made up 37 percent compared to almost 60 percent on the GPT-4O predecessor.

With GPT-4.5, Openai Continues to bet on big, expensive big linguistic models, despite the appearance of high -capable smaller foods such as Chinese Deapeek’s R1 startup, which are open, cheaper and more affordable for developers.

It comes when the competition grows within the rapidly developing industry AI, as in recent weeks, technological groups have rushed to run their latest models. On Monday, Anthropic revealed its sonnet Claude 3.7, which took place last week launch Grok 3, the latest XAI MAI model.

Openai in a blog report on Thursday stated that the GPT-4.5 had “broader knowledge and a more profound understanding of the world, which led to a decrease in hallucinations and greater reliability in a wide range of topics.”

“With each new order of computing, new opportunities are coming,” the company said, adding that the GPT-4.5 was “on the verge of what is possible in an uncontrolled training.”

Openai stands on the avant -garde of the global race to head the II industry by gathering tens of billions from investors to finance large models with increasing opportunities that require a huge amount of computing.

In conversations with Softbank and other investors to raise up to $ 40 billion when evaluating $ 300 billionincluding new money. Anthropic also raises funds about $ 3.5 billion when estimating $ 60 billion, two people who know the process have stated.

However, huge expenses for the management of large models forced Openai to consider removing the developer access to the GPT-4.5 against the background of fierce competition at competitors.

Openai said that at a time when the developers who pay for the use of Openai models will be available preview GPT-4.5, this access can be canceled in the future.

AI groups greatly bring profits through paid API access and individual subscriptions. Openai said he would see how the developers use a powerful model and whether they should be offered them, given the high cost of its launch.

The company said: “GPT-4.5 is a very large and intense model, making it more expensive than not replacement for the GPT-4O (its predecessor). Because of this, we estimate whether it should continue to serve it in the long run, as we balance the support of modern models.”

Earlier, Openai Sam Altman Executive Director said GPT-4 costs more than $ 100 million for training, and such costs are supposed to increase as the size and possibility of scale models and require more computing for training and launch.

In a report on X after the announcement Altman said the company was “by graphic processors”, the chips required to launch and training AI Systems.

“This is not how we want to work, but it is difficult to predict the growth that leads to lack of schedules,” he said, adding that they are expecting more in the coming weeks.

He also emphasized that the GPT-4.5 was not focused on the reasoning and would not beat industry landmarks, but he had “a different kind of intelligence, and there is a magic I had never felt.”

Additional George Hamond Report in San Francisco and Melissa Hakekil in London