Openai reveals GPT-4.5 amid a wave of new artificial intelligence model versions

Digest opened free editor

Rola Khaleda, FT editor, chooses her favorite stories in this weekly newsletter.

Openai launched the largest artificial intelligence model with fewer “hallucinations”, the latest in a wave of AI’s versions by American technology groups.

The San Francisco-based company revealed on Thursday the GPT-4.5, its long-awaited update for technology that supports its famous product. In early tests, hallucinations, where Amnesty International The systems generate inaccurate information, 37 percent were compared to nearly 60 percent on the GPT-4O predecessor.

With GPT-4.5, Openai It continues to bet on the large and expensive language models, despite the appearance of very capable smaller products, such as the Chinese R1 R1 R1, which are open, cheaper and more accessible to developers.

This comes with the growth of competition within the artificial intelligence industry rapid development, as technology groups have rushed to launch their latest models in recent weeks. Anthropor unveiled Claude 3.7 Sonnet on Monday, which followed the Grok 3 launch last week, the latest model of Elon Musk’s Xai.

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

The company said: “With every new arrangement the size of the new capacity account,” adding GPT-4.5 was “on the limits of what is possible in learning not subject to supervision.”

Openai was at the forefront of a global race to lead the artificial intelligence industry, raising tens of billions of investors to finance larger models with an increase in capabilities that require huge amounts of computing power.

Talks are with SoftBank and other investors to raise up to $ 40 billion in evaluation 300 billion dollarsIncluding new money. Two people were aware of this process, that Antarbur raised donations of about $ 3.5 billion, with a value of $ 60 billion.

However, the huge costs of operating larger models prompted Openai to consider the developer’s access to GPT-4.5 amid fierce competition for competitors.

OpenAi said that while a preview of GPT-4.5 will be provided for developers who pay for the use of Openai models through the API programming interface, this access can be canceled in the future.

Artificial intelligence groups are largely generated by revenues by the arrival of the paid application programming interface and individual subscriptions. Openai said it will see how developers use the strong model and whether it is worth it to give them the high cost of its operation.

The company said: “GPT-4.5 is a very large and intense account, which makes it more expensive than and not an alternative to GPT-4O (its predecessor). For this reason, we evaluate whether we will continue to provide it in the long term API where we balance the current capacity support with the construction of future models.”

The CEO of Openai Sam Altman said that GPT-4 costs more than $ 100 million for training, and these costs are expected to increase widely with the size and capabilities of models and require more computing power of training and employment.

In a post on X after the announcement, Altman said that the company was “outside the graphics processing units”, and the chips required to operate and train artificial intelligence systems.

“This is not how we want to work, but it is completely difficult to increase the growth that leads to a lack of graphics processing unit,” he said, adding that they expect more in the coming weeks.

The GPT-4.5 also highlighted not focused on thinking and will not overcome the standards of industry, but it was “a different kind of intelligence and there is a magic that I have not felt before.”

Participated in additional reports of George Hammond in San Francisco and Melissa Haykal in London