Microsoft Will Unveil Its Own GitHub Copilot Coding Model at Build — a Direct Shot at Claude Code

Microsoft is preparing to unveil a homegrown coding model built to power GitHub Copilot at its Build developer conference in San Francisco on June 2–3, according to reporting from The Information. The coding model is the headline of a broader slate of in-house "MAI" models the company is expected to show, and it marks Microsoft’s most direct attempt yet to put its own intelligence underneath one of its most important developer products.

The coding model arrives alongside a refreshed family of MAI systems spanning speech, voice and imagery — including a text-to-image model and multilingual transcription and text-to-speech successors — destined for products such as Copilot, Teams and Azure. Taken together, the lineup is meant to reduce Microsoft’s dependence on outside model providers, namely OpenAI, Anthropic and Google, by giving the company silicon-to-software control over the models its flagship apps run on.

The effort is led by Microsoft AI chief executive Mustafa Suleyman, whose internal division had been restricted from training top-tier frontier models under the terms of Microsoft’s OpenAI partnership. Those restrictions were renegotiated in April, clearing the way for Suleyman’s team to build competitive models of its own — and Build is the stage where that pivot becomes public. The message to developers, customers and investors is that Microsoft wants more ownership of the model layer, not just a reseller’s seat on top of OpenAI.

The competitive subtext is hard to miss. GitHub Copilot was the original AI coding assistant and long the default, but it has steadily lost mindshare to Anthropic’s Claude Code, which many developers now reach for first. A Microsoft-built coding model gives the company a way to tune Copilot’s core engine to its own roadmap and economics rather than paying to route every request to a rival’s API — a calculus that matters even more as GitHub shifts Copilot to usage-based, token-metered billing this month.

Microsoft has not confirmed final names, benchmarks or availability for the coding model ahead of the keynote, and a Build demo is not the same as general availability. But the strategic direction is unambiguous: after years of leaning on OpenAI’s models, Microsoft is signaling that the next chapter of Copilot will increasingly run on intelligence it builds and controls in-house.