• Microsoft is reportedly working on its own in-house generative AI models and is planning to release them to customers.
• The tech giant invested billions into ChatGPT-maker OpenAI in order to use its models but now seems to be planning to move away from the company.
• Microsoft reportedly has three reasoning AI models in the pipeline under the name MAI.

A new report details that Microsoft is working on developing its own AI models that could compete with those from OpenAI, and it is looking to sell them to developers in a bid to continue reducing its dependence on the ChatGPT-maker on which it invested around $13 billion.

Currently, Microsoft uses OpenAI’s models in its Copilot software, a major piece of its latest Microsoft 365 offering, however, it reportedly has been testing models from xAI, Meta and upstart DeepSeek as potential replacements for OpenAI in Copilot, alongside its own models being worked on.

One of the major selling points for Copilot us that it uses ChatGPT’s GPT-4 model, believed to be among the most advanced and powerful in the world.

It believes reducing its dependency on OpenAI will allow it to diversify its technology from what OpenAI is making and reduce costs.

Per The Information, one of the names it is testing; China’s DeepSeek, shook the tech world when it announced earlier this year new AI models that could compete with the processing power of those from OpenAI but at a fraction of the cost.

But as the generative AI race heats up with new entries and new technologies, the scramble for a share of the profits has pushed the big tech companies that can afford it to develop in-house. Microsoft’s own AI models are currently being tested under the original name “MAI.”

There are apparently three reasoning models under this name that are currently being trained. These models reportedly use “chain-of-thought” techniques. This process simulates reasoning when tasked to solve complex problems.

“Just as humans use a single brain for both quick responses and deep reflection, we believe reasoning should be an integrated capability of frontier models rather than a separate model entirely. This unified approach also creates a more seamless experience for users,” it added in terms of why hybrid reasoning was implemented here,” Anthropic said of its Claude 3.7 Sonnet hybrid reasoning model.

“Reasoning” in generative AI is the latest innovation in the technology itself and allows improved performance on math, physics, coding and other advanced tasks. Reasoning is a step towards what Big Tech is selling as the future of generative AI: AI agents.

These pieces of AI-powered software will be able to perform tasks for users either on their PCs or on their smartphones, like composing and sending emails for them, or booking and paying for plane tickets.

As for Microsoft, the report says that the company is planning to release its MAI models later this year as a way to let developers use them in their own apps, perhaps similarly to what DeepSeek allows but obviously gated by a subscription to Microsoft 365.