Microsoft Plans to Release a New Proprietary Programming Model Next Week

According to The Information, Microsoft is planning to unveil a suite of proprietary AI models, including a coding model specifically tailored for programming scenarios, at its annual Build developer conference in San Francisco next week, aiming to enhance GitHub Copilot's competitiveness in the market.

GitHub Copilot was once a pioneer in the AI programming assistant race, but its early advantage has been gradually eroded by Cursor and Claude Code. The introduction of this proprietary coding model represents an important step for Microsoft in attempting to regain its developer market share. In addition to the coding model, Microsoft also intends to release a series of models of different sizes that cover specialties such as transcription, reasoning, speech, and images.

This release is significant for Microsoft. Since the AI chief Mustafa Suleyman joined the company two years ago, Microsoft's proprietary models have never topped AI capability rankings. Suleyman previously stated that due to the cooperation agreement with OpenAI, Microsoft could not independently pursue human-level AI research and development until the renegotiation in April this year.

Currently, Microsoft mainly relies on OpenAI (which is provided for free through 2032 according to the agreement) and Anthropic (purchasing models) to power GitHub Copilot and the 365 Copilot in the Office suite. Dependence on the Anthropic models has forced Microsoft to increase the prices of some products and set limits on developers' usage. If the proprietary coding model can become a low-cost alternative to Claude, it will help Microsoft win back developer customers who are sensitive to costs.

Microsoft will price the new models below those of OpenAI and Anthropic's similar products and provide access to developers through the Azure cloud platform, consistent with the way OpenAI and Anthropic models are purchased. In the long run, when Microsoft loses its free access to OpenAI technology in about six years, having mature in-house AI capabilities will be key to breaking free from dependence on external suppliers.