OpenAI’s GPT-5 is gaining where it matters most: Enterprise

Sam Altman turned Openai into a cultural phenomenon with Chatgpt.

Now, three years later, he follows where the real money is: Enterprise.

Rocky was the release of Openai’s newest artificial intelligence model last week, GPT-5. Critics suppressed the less intuitive feeling, and ultimately led the company to restore the inheritance GPT-4 to Chatbot customers.

However, the GPT-5 is not related to the consumer. Openai’s effort to break the corporate market in which its rival Antropic started.

Within a week and interventions such as cursor, vercel and factory, they have already made the GPT-5 as the default model in some key products and vehicles, faster installation, making better results and lower price in complex tasks.

Some companies said that the GPT-5 had matched or defeated with Claude in a space anthropic code and interface design, which was once dominated.

BoxAnother corporate customer tests the GPT-5 in long, logic-extensive documents. CEO Aaron Levie said to CNBC that the model was a “breakthrough” and performed with a reasoning level that previous systems could not match.

Behind the curtain, Openai established its own corporate sales team – more than 500 people under Coo Brad Lightcap. MicrosoftThe chief investor and key cloud partner of the initiative. Customers can access GPT models via Microsoft Azure, or go directly to Openai, which controls API and product experience.

Still, the economy is cruel. It is expensive to run models and they spend great to lock both openai and anthropic customers, Openai is on the road to burn $ 8 billion this year.

The cursor, which is still a large anthropic customer, directs new users to Openai. Michael Truell, the founding partner and CEO of the company, underlined the change OpenAi’s launch live flowTo describe the GPT-5 as the “smartest coding model we’ve ever tried”.

Truell, existing cursor customers will continue to use anthropic as the default models, the change is valid only for new records, he said. The cursor continues a decisive income contract with anthropic, who built his business to dominate the corporate layer.

As of June, the Enterprise income constitutes approximately 80%, annual income increased by 17 times annually, said that there is a person who knows the issue that requires anonymity to discuss company data. The company added 3 billion dollars in the last six months, including 1 billion dollars in June, and said that this year has already signed the number of eight and nine -digit agreements this year compared to the entire 2024.

Antropic said that the institutional footprint extends far beyond technology.

POWERS Tools for Claude Amazon Asal alexa and AIGAnd the drug is used by the best players in retail, aviation and professional services. The company is embedded in Amazon Web Services, GCP. SnowflakeDatabricks and Palantir – And agreements tend to expand quickly.

The person said that average customer expenditures grew more than five times last year and more than half of business customers now use more than one Claude products.

The person said that the revenue for the rest of the work has grown more than one annually, except for the two largest customers.

Even with this wide access, Openai gains the ground with corporate customers.

The use of GPT-5 API has increased since the launch, the model is now a lot of coding and creating agents, and a person who knows the issue that requires anonymity to discuss company data, cases of reasoning use of more than eight times.

Especially for planning and multi -stage reasoning tasks, corporate demand is increasing sharply.

Coding Platform that works with AI Qodo has been tested soon GPT-5 Gemini 2.5, including Claude Sonnet 4 and Gok 4, including high-level models and in a blog post leads to capture coding errors, he said.

The company was the only model that caught critical problems such as security errors or broken code, suggested clean, focused corrections and proposing the code that did not need to change. Weaknesses occasionally contained the wrong positive and some surplus.

Vercel, a cloud platform for web applications, made the GPT-5. default The new open source “Vibe Coding” platform-A system that transforms all English demands into live, working applications. He also rolled the GPT-5 into the garbage agent, which he said was particularly good in balancing complex tasks and thinking with long instructions.

“While there is already a lot of competition in AI models, Claude had only this area. It was a very good coding model. It wasn’t even close.” He said. “Openai wasn’t just in the game.”

This has changed with GPT-5.

“At least they were caught,” Ubl said. “Better in some things, worse in other things.”

The GPT-5 said that early-stage came forward for prototyping and product design and that Claude was more creative than the sonnet.

“Traditionally, you should optimize for the new model, and we’ve seen really good results from the beginning,” he said.

Jetbrains adopted GPT-5 By default in the AI assistant and kineto, a new coded tool to create websites and applications can produce faster, single -purpose tools than user requests. Developer platform Factory said He collaborated with Openai to make GPT-5 default for his vehicles.

“GPT-5 is much better when it comes to getting a really good plan to implement a complex coding solution. “It is much better to plan his plan for a long time and to have consistency for a long time.”

Grinberg added that GPT-5 has combined well with very agent platforms: “Low-level application plays with most of these high-level details that we manage.”

Pricing flexibility was an important factor in the factory’s GPT-5 decision.

Pricı The pricing is often what our end users care about, Gr Grinberg said that cheaper inference is now making customers more comfortable attempts. Instead of guessing whether a problem is worth a cost, they can “shoot more easily than the hip” and discover ideas without hesitation.

Anton Osika, Founding Partner of Lovable, CEO, said that an AI-backed vehicle, which allows everyone to create real software businesses without writing a single code line, said that their team tested GPT-5 for weeks before being officially released and said that it was “super happy” with improvement.

“What we find is more powerful. In the case of many complex use, Smart is smarter, Os Osika said, the new model is more prone to action and reflecting the action it takes, and” spends more time to ensure that it is really true. “

Box‘S Levie said that the biggest gains for him emerged in corporate workflows that are not related to the writing code. The team has been testing the model for weeks about real world job data, from a hundred -page lease contracts to product road maps, and found that it was successful in problems that wearing previous AI systems.

Levie added that this step change improvements are critical for the institutional use in which AI agents work in the background to carry out tasks, and that the GPT-5 can make a real breakthrough for business automation.

“The GPT-5 has performed incredibly well-the best model of Openai-and the best available in most of our tests,” he said.

– Kevin Schmidt from CNBC contributed to this report.

Watch: Openai started the GPT-5 model