Sam Altman transformed Openai into a cultural phenomenon with Chatgpt.
Now, three years later, he continues where the real money is: the company.
The deployment last week of GPT-5, the new OpenAi artificial intelligence model, was Rocky. Critics struck its less intuitive feeling, which finally led the company to restore its GPT-4 inheritance to Chatbot customers.
But GPT-5 does not concern the consumer. It is OpenAi’s effort to break the business market, where Anthropic Rival was one step ahead.
A week later, and startups like Cursor, Vercel and Factory say they have already made GPT-5 the default model in certain key products and tools, praising its faster configuration, better results on complex tasks and a lower price.
Some companies have declared that GPT-5 now corresponds or beats Claude on the code and interface design, an formerly dominated anthropogenic space.
BoxAnother business client tested GPT-5 on long and heavy documents. CEO Aaron Levie told CNBC that the model is a “breakthrough”, saying that it reaches a level of reasoning that previous systems could not match.
Behind the scenes, Openai has formed its own business sales team – more than 500 people under the head of the Brad Lightcap – operating operating independently of Microsoftwho was the main investor of the startup and the Cloud key partner. Customers can access GPT models via Microsoft Azure or go directly to Openai, which controls the API and product experience.
However, the economy is brutal. The models are expensive to execute, and Openai and Anthropic spend great to lock customers, Openai on the right track to burn $ 8 billion this year.
Win the business
The GPT -5 is significantly cheaper than Opus 4.1 from Claude Top -End of Anthropic – by a factor of seven and a half years, in some cases – but Openai spends huge quantities in infrastructure to support this advantage.
For Openai, it is a pressure to win customers now, to have them locked up and create a real business on the back of this loyalty.
Cursor, still a major anthropogenic client, now directs new users in Openai. The co-founder and CEO of the company Michael Truell underlined the change during Livestream launching of OpenaiDescribing GPT-5 as “the most intelligent coding model we have ever tried”.
Truell said that the modification only applies to new registrations, because existing cursor customers will continue to use Anthropic as default model. Cursor maintains a income contract committed with Anthropic, which has developed its activities on the domination of the corporate layer.
In June, the company represented around 80% of its income, with income from 17 times on the other, said a person familiar with the question that had requested anonymity in order to discuss the company’s data. The company added $ 3 billion in revenue in the last six months – including $ 1 billion in June only – and has already signed the triple in the number of eight and nine figures this year compared to all 2024, said the person.
Anthropic said that his corporate imprint extends far beyond technology.
Claude Powers Tools for Amazon Prime, Alexa, and AigAnd is used by the best players in pharmaceutical, commercial, aeronautical and professional services. The company is integrated on Amazon Web Services, GCP, SnowflakeDatabricks, and Palantant – And its transactions tend to develop rapidly.
Average customer expenses have increased over five years in the past year, with more than half of commercial customers now using several Claude products, the person said.
Excluding its two largest customers, revenues for the rest of the company have increased more than eleven times on the other, the person said.
Even with this wide range, Openai is gaining ground with business customers.
The use of the GPT-5 API has increased since its launch, the model now dealing with more than twice as much coding and agent work, and the user reasoning that jumped more than eight, said a person familiar with the question that had requested anonymity in order to discuss the company’s data.
Business demand increases sharply, in particular for planning and reasoning tasks in several stages.
GPT-5 improvement
The traction of the GPT-5 during last week shows how speed can change when performance and prices in favor of the Openai.
Coding platform fueled by AI Qodo recently tested GPT-5 against high-level models, including Gemini 2.5, Claude Sonnet 4 and Grok 4, and said in a blog article he had led to catch coding errors.
The model was often the only one to catch critical problems, such as safety bugs or broken code, suggesting clean and targeted fixes and jumping from the code that didn’t need to change, the company said. The weaknesses included occasional false positives and a certain redundancy.
Vercel, a cloud platform for web applications, has made GPT-5 default In its new open-source “ambient” coding platform-a system that transforms the invites into English into live and work applications. He also launched GPT-5 in his dashboard agent, where the company said that she was particularly good for juggling complex tasks and thinking about long instructions.
“Although there has already been a lot of competition in AI models, Claude only had this space. It was by far the best coding model. It was not even close,” said Malta Ubl, CTO of Vercel. “Openai was simply not in the game.”
It changed with GPT-5.
“They have at least caught up,” said Ubl. “They are better in certain things, they are worse in other things.”
He said that the GPT-5 stood out for prototyping and product design at an early stage, calling it more creative than Claude’s sonnet.
“Traditionally, you need to optimize for the new model, and we have seen very good results from the start,” he said about the ease of integration.
Jetbrains adopted GPT-5 As a defect in its AI assistant and in Kineto, a new tool without code to create websites and applications, after finding that it could generate simple and single -use tools faster from user prompts. Developer platform The factory said He collaborated in close collaboration with OpenAi to make GPT-5 the default value of his tools.
“When it comes to obtaining a very good plan for the implementation of a complex coding solution, GPT-5 is much better,” said Matan Grinberg, CEO of Factory. “It is much better to plan and have a consistency on its plan over a long period.”
Grinberg has added that GPT-5 integrates well in their multi-agent platform: “He plays very well with lots of these high-level details that we manage at the same time as the details of low level implementation.”
Price flexibility has also been a major factor in the decision of the GPT-5 defect faking plant.
“The price is mainly what our end users are careful,” said Grinberg, adding that this cheaper inference now makes customers more comfortable. Instead of guessing if a question is worth it, they can “pull the hip more easily” and explore ideas without hesitation.
Anton Osika, co-founder and CEO of Lovable, a company that builds a tool powered by AI which allows anyone to create real software companies without writing a single line of code, said that his team was testing the beta version of GPT-5 for weeks before its officially launch and was “super satisfied” with improvement.
“What we have found is that it is more powerful. It is more intelligent in many complex cases of use,” said Osika, adding that the new model is “more inclined to take measures and to think about the action it takes” and “devote more time to ensure that it really gets things”.
BoxLevie said that the greatest gains for him appeared in corporate workflows that have nothing to do with code writing. His team has been testing the model for weeks on complex and real commercial data – hundred pages rental agreements with products’ roadmaps – and found that it excels in problems that have stumbled earlier the AI systems.
Levie has added that for the use of the company, where AI agents work in the background to perform tasks, these step change improvements are essential and can transform the GPT-5 into real pierced for the automation of work.
“GPT-5 has perceived incredibly good-certainly the best model of Openai-and in many of our tests, this is the best available,” he said.
– Kevin Schmidt of CNBC contributed to this report.
Watch: Openai launches the GPT-5 model
