Microsoft (MSFT.US) accelerates AI autonomy and self-developed image generation models will replace OpenAI related technology

Mustafa Suleyman, head of Microsoft's AI business, said in an interview that compared to the OpenAI model, Microsoft's self-developed model runs faster, costs less, has higher generation quality, and can also improve user retention.

In recent years, Microsoft has been developing large models that can compete with OpenAI and Anthropic, hoping to gradually integrate AI capabilities into its product systems and reduce AI service costs. Suleyman revealed that taking PowerPoint as an example, the operating cost of Microsoft's self-developed Microsoft AI (MAI) model was reduced by about 85% compared to the previous one.

Although according to the long-term cooperation agreement between the two parties, Microsoft can still use OpenAI's large models for free. Microsoft needs to bear the large-scale computing power infrastructure costs required to run these models, so adopting a more efficient self-developed model is expected to bring considerable cost savings.

Suleyman said that in office software such as PowerPoint, users often use AI to generate and edit images, such as making infographics, etc., and image generation is also one of the most commonly used AI functions by new users, so improving image generation quality and response speed is essential to improving the user experience.

In addition to image generation, Microsoft is also speeding up the promotion of other self-developed AI models. The company announced that companies including T-Mobile US (TMUS.US) and the British low-cost airline easyJet have begun deploying Microsoft's latest AI voice generation model in customer service call centers.

Earlier, there were media reports that Microsoft has begun gradually replacing OpenAI and Anthropic's text generation models with self-developed models in various Office software. Microsoft said that the performance of its MAI model for common tasks in applications such as Excel is already comparable to OpenAI's latest GPT-5.6 model, but it is more cost efficient.

Suleyman said that currently the MAI model has been applied to more than half of Microsoft's products, and tests have been carried out on all Microsoft products. He said, “This is a very strong start, and we will continue to do our best.”