Microsoft Develops Its Own AI Models to Compete with OpenAI and Anthropic

Microsoft is working hard to build its own family of AI language models, led by Mustafa Suleiman. The goal is to match the capabilities offered by industry leaders like OpenAI and Anthropic. According to The Information, Microsoft plans to make its models available via API by the end of 2025.

The family of models under development, known as MAI, is already showing results close to those of leading solutions from OpenAI and Anthropic on standard benchmarks. In particular, one of the models, designed for inference, is designed to compete with OpenAI’s flagship model, GPT-4.

Unlike Microsoft's earlier Phi-series models, which focused on balancing cost and performance, the new MAI models are significantly larger in scale and power, allowing them to tackle more complex workloads and compete with the best offerings on the market.

Microsoft is not limiting itself to internal developments. The company is already testing models from other leading players, such as xAI, Meta* and DeepSeek, as part of its Copilot service. This could be part of a strategy to reduce its reliance on OpenAI, despite the close partnership between the two companies.

The planned release of APIs to third-party developers by the end of 2025 will allow Microsoft to compete directly with OpenAI, Anthropic and other AI labs offering similar services.

Developing its own AI models and integrating them into the Microsoft ecosystem opens up new opportunities for the company. This not only strengthens its position in the AI ​​market, but also allows it to offer customers more flexible and competitive solutions.