Openai’s latest model will change the economics of software

Nevertheless, a point of consensus has surfaced since then. The model, as well as its predecessor, O1 (O2 is left because it is the name of a European mobile network), the more “thinking” does in response to a sign gives the more better results. More thinking means more computing power – and more cost per querry. As a result, there is a big change in the economics of the digital economy that provides cheap services to a large number of people on low marginal costs, the credit goes to free distribution on the Internet, to query, zero-seam-dlag era and more Has been left behind.
Investors give Openai like a technical beloved: Recently it is priced at $ 157bn on the basis of funding. He hopes that it will become the next trillion-dollar company thanks to the success of products like ChatjPT. But the high cost of state-of-the-art models, as well as other pressures of suppliers, distributors and competitors suggest that model-building cannot provide powers such as monopoly received by large technology. “A very important thing to understand about the future: AI’s economics is completely changing,” an AI researcher Francois Chollet said the day of O3 on the social-media site X.
Mr. Cholat has helped to encourage the excitement about O3. In June, he launched a $ 1 million prize for models that he could fulfill the challenge called “Abstraction and Reasoning Corpus” or ARC. It is a pile of simple looking visual-vandal riddles (see diagrams). To be “easy for humans and impossible for modern AI”. This award was not only challenging for itself. Mr. Cholat said that completing ARC work is a “important” step towards the creation of artificial general intelligence, which means that machines beat humans in many tasks.
Six months later, OpenAI succeeded in the test. Its O3 model scored a breakthrough score of 91.5%. Mr. Cholat said that its success in the challenge showed AI a step in the ability to customize new tasks. The new model is not only better; it is different. Like the O1, it uses the “Test-Time Compute” approach, the more time it costs to estimate (when a trained AI model answers questions) gives the more better results. , O3, in fact, is designed to think more deeply about the question.
It costs more here. Mr. Cholat has set a limit of $ 10,000 on the amount that contestants can spend on computing power to answer 400 questions in their challenge. When Openai put forward a model under the border, he spent $ 6,677 (about $ 17 per question) to score 82.8%. A score of 91.5% obtained by O3 came from flying the budget. The company did not disclose the amount spent, but said that in the expensive version of the process, the quantity of “calculation” has been used as a cheap approach – suggested about $ 3,000 to solve a question. Gone in which man takes a few seconds.
The previous AI model had already challenged the low-b-wing-level criteria of the software industry, as to answer the questions required a much higher processing power than using an equivalent tool such as the search engine. But the cost of creating and running big language models was so low that OpenAI could still provide free access.
This is no longer so with the latest models. OpenAI O1 limits the “Pro” version of the “Pro” version to the users at the membership level (and according to its owner Sam Altman, loses the money, as the customers are spending more than the budget prescribed by the company. ). Piergu of New Street Research, a company of analysts, believes that OpenAI can charge up to $ 2,000 per month for full access to O3.
The power of such models depends on bringing a version of the “scaling laws” of the sector to the last user. Till now, progress in AI was dependent on large and better training runs, in which more data and more computer power used to produce more intelligence. But once trained a model, it was difficult to use additional processing power well, as the ARC challenge shows the success of O3, it seems that the scaling law now estimates the training model now Have not increased.
Such growth changes the economics that come forward to model-makers such as OpenAI. Dependence on more processing power strengthens their suppliers, such as Nvidia, who is the manufacturer of expert AI chips. This also benefits the distributors of the AI model, especially cloud-service providers such as Amazon, Microsoft and Alphabet. And this can justify the fate of technical giants that continue to invest in data centers because more estimates will require more computing power. On January 21, Mr. Trump announced “Stargate“, A large private sector project to create data centers in the US, incorporating Opai. The firm is being pressed on both sides.
Then there is competition. Google has released its own region model, Gemini 2.0 flash, and possibly other technical companies will also release. It is expected to follow the open-source model. Customers will be able to get many models from various providers. And although generative-AI models can improve slightly through their interactions with customers, they lack real network effects, unlike Google and Facebook products created in the previous era.
High marginal cost means that model-bedters will have to generate meaningful prices to collect premium prices. LAN Guan of Consultancy Extense says that it is expected that models like O3 will support AI agents that individuals and companies will use to increase their productivity. Even a high price for the use of logic models, a fully developed mathematics PhD, may be worth it than the cost of hiring. But it depends on how useful models are.
Cases of different use can also cause more fragmentation. Jeremy Schneider of Consultancy McKinse says that corporate customers will require models to provide AI services that are specific to the needs of each enterprise rather than a general-purpose model like chat GPT.
Instead of dominance of a firm, some people expect the model -manufacture that be a oligatous, which will have high obstacles for entry but will have no obstacles – or there will be no monopoly profit. For now, Openai is the leading, but one of its main rivals, anthropic, allegedly raising funds on an $ 60 billion evaluation, and Elon Musk’s majority with the majority XAI is worth $ 45 billion. This shows that they also have high expectations from them. Openai has shown its technical lead with O3, but its business model is still unused.
© 2025, The Economist Newspaper Limited. All rights reserved. From The Economist, published under license. The original material can be found on www.economist.com
(Tagstootransite) OpenAI (T) Artificial Intelligence (T) AI (T) OpenAI O3
#Openais #latest #model #change #economics #software