Microsoft will officially release a number of self-developed artificial intelligence models at Build, its annual developer conference in San Francisco next week. This move is intended to gain recognition from the developer community. According to people familiar with the plans, Microsoft plans to launch a code-specific model to enhance the competitiveness of its code assistant GitHub Copilot. This tool once had a first-mover advantage in the field of AI code assistance, but now its market share is being eroded by Cursor and Claude Code.
The person familiar with the matter also said that Microsoft will also release a number of new models with different parameter specifications, specializing in scenarios such as speech transcription, logical reasoning, speech processing, and image generation. This series of new models is an iterative upgrade based on the self-developed model Microsoft showed earlier this year. As Stephanie explained in detail in Wednesday's report, speech processing and speech transcription functions have become a popular direction for developers.
An official Microsoft spokesperson declined to comment. Mustafa Suleiman, head of Microsoft's artificial intelligence business, could no longer hold back and on Tuesday revealed an upcoming image generation model in advance.
This new product release is a key test of Microsoft’s self-developed AI technology strength. In the two years since Suleiman joined the company, the various models launched by his team have never topped the global AI capability rankings. Suleiman once admitted that due to the restrictions of the cooperation agreement between Microsoft and OpenAI, the company had been unable to develop top-level models. The agreement stipulated that Microsoft was not allowed to develop AI that reached the level of human intelligence before the two parties renegotiated the terms in April this year.
For a long time, Microsoft's GitHub Copilot and Microsoft 365 Copilot, which can automate tasks for office software, have mainly relied on OpenAI and Anthropic models to run. Among them, Microsoft can use the OpenAI model for free, and the cooperation rights will last until 2032; using the Anthropic model requires a fee.
In about six years, Microsoft will no longer have access to OpenAI’s technology. To this end, the company hopes to build a mature self-owned AI system and get rid of its dependence on high-priced external technology suppliers.
People familiar with the matter said that Microsoft will focus on the new model's high cost performance, and the price will be lower than similar products with slightly better performance than OpenAI and Anthropic. This positioning idea continues Microsoft’s previous marketing strategies for many self-developed models, and is also similar to Google’s layout strategy. Of course, the actual performance of the new model still needs to be verified by developers, and Microsoft's propaganda cannot be relied upon.
Microsoft has previously stated that some AI functions of the office software Copilot have been integrated into self-developed models, effectively reducing operating costs. According to people familiar with the matter, the new model launched this time will also further play a role in reducing costs.
Microsoft also hopes that developers can purchase these new self-developed models through its Azure cloud platform just like they purchase OpenAI and Anthropic model services.
Due to its reliance on Anthropic's technology, Microsoft has increased the prices of some products such as GitHub Copilot, and also limited the amount of developers who can call Anthropic models in the tool. If Microsoft's self-developed code model can become an affordable alternative to Claude, it will help it win back cost-sensitive developer users and regain market share.