Microsoft is planning to announce multiple in-house AI models at its annual Build developer conference in San Francisco next week. Chief among them is a custom coding model designed to give GitHub Copilot a much-needed boost against rivals like Cursor and Claude Code, which have been eating into its market share.
Beyond the coding model, Microsoft is also prepping a range of large language models with various parameter counts for transcription, reasoning, speech, and image tasks. This will be a major test for the R&D team led by Microsoft AI CEO Mustafa Suleyman, which hasn't yet produced a top-ranked model. Previously, Microsoft's self-development was constrained by an exclusive agreement with OpenAI that limited the company from building human-level AI until April 2026. But the two companies revised their partnership terms this April, giving Microsoft back its freedom to innovate.
Currently, Microsoft relies on free access to OpenAI's technology (licensed through 2032) and paid Anthropic models to power GitHub Copilot and Microsoft 365 Copilot. The high cost of external models has already forced Microsoft to raise prices on some GitHub Copilot plans and cap usage of Anthropic models. By distributing cheaper in-house models through Azure, Microsoft can offer developers a low-cost alternative to Claude and reduce the operating costs of AI features in Office apps. More importantly, with the OpenAI free license expiring in about six years, these proprietary models will be Microsoft's critical leverage to break free from expensive suppliers and maintain its independence.