Microsoft Relying heavily on Openai’s artificial intelligence model to power AI capabilities in its key products. It tries to reduce this dependency.

The software company said Thursday it has begun publicly testing native AI models that could enhance its vice assistant consumers. The MAI-1-preiview model is being tested on LMARENA, a website that can be evaluated.

“In the next few weeks, we will roll out MAI-1-preiview in some text use cases within Copilot to learn and improve from user feedback,” Microsoft said in A. Blog Posts. The company has published a form that developers can request early access.

Meanwhile, Microsoft remains a key supporter of Openai and a strategic partner of AI startups Pay attention to About $500 billion. Microsoft has invested more than $13 billion in OpenAI, which in turn relies on Microsoft’s cloud infrastructure to run its models.

Microsoft borrows models of OpenAI-to-Power functionality from Bing, Windows 11 operating systems and other products.

But these companies are clearly entering the competition in various ways. Last year, Microsoft additional Openai access to competitors list annual report. For years, this is a lineup that includes Megacap peers Amazon, , , , , apple, , , , , Google and Yuan.

In recent months, Openai has turned to Other cloud providersFor example CoreWeaveGoogle and Oraclemeets a large number of needs. Chatgpt Assistant for the startup is now here 700 million people One week.

On LMARENA, Microsoft’s new model ranks 13th in Thursday’s text workloads, and below are models from Anthropic, DeepSeek, Google, Mistral, OpenAI and XAI.

Microsoft said in its blog post that the model is perfected with the help of about 15,000 Nvidia’s H100 graphics processing unit, it also has a working group of NVIDIA GB200 chips.

“We have great ambitions for what we’re going to do next – advances in models, exciting algorithms and the opportunity to attract billions of people through Microsoft’s products,” said Mustafa Suleyman, CEO of Microsoft’s AI division. postal On X.

Microsoft has previously developed a small open source language model under the name of PHI.

Suleiman wrote on X that Mai-1-preview stands for “end to end of our first basic model training.”

Suleyman once competed with OpenAI at the corner of a startup. last year, Microsoft hired him and many of his inflection point colleagues. Before that, Suleyman Google Purchased in 2014.

Suleyman’s group inside Microsoft has been expanding, about Twenty people In recent months, it comes from Google’s DeepMind AI Lab.

watch: Openai CFO Sarah Friar: The biggest problem we face is “continuously being calculated”