Microsoft builds its own AI models — and that's good news for your costs.

What Microsoft announced

At its Build developer conference in San Francisco on 2 June 2026, Microsoft announced MAI-Code-1-Flash — its first proprietary AI model, focused on code generation and developer tasks. Alongside it, Microsoft introduced MAI-Thinking-1, a reasoning model designed for high accuracy at a low token cost.

Both models are intended to reduce Microsoft's dependence on OpenAI's GPT family of models, which have powered Copilot, Azure AI, and a range of Microsoft 365 integrations since 2023. By owning its own model layer, Microsoft can control costs, set pricing independently, and respond faster to competitive pressure from Anthropic, Google, and Meta.

Microsoft stated that the new models would initially be available to Azure developers via API, with integration into Copilot products to follow as the models mature.

Why this affects businesses using AI tools

The AI tools most businesses use — Copilot in Microsoft 365, ChatGPT, Claude, Gemini, and the dozens of workflow automation tools built on top of these — are priced partly based on the underlying model costs their providers pay. When that underlying market becomes more competitive, the pricing pressure flows downstream.

Three dynamics are now visible in the AI infrastructure market:

• More supply: Microsoft, Anthropic, Google, Meta, and Mistral are all shipping capable models simultaneously. Supply of model capacity is rising faster than demand.
• Price competition: OpenAI cut API prices for GPT-4o by more than 50% in 2025. Google cut Gemini API pricing shortly after. More competition from Microsoft's own models will sustain that pressure.
• Efficiency improvements: MAI-Thinking-1 is specifically designed to deliver reasoning performance at lower token cost — meaning the same output quality for less compute spend.

For a business currently paying for AI-assisted tools — CRM automation, email drafting, quote generation, scheduling — the combination of falling model costs and rising model capability means the return on those tools will improve over the next 12 months without you needing to do anything. And for businesses that have not yet started, the cost barrier is lower now than it was six months ago.

What to watch

Microsoft's Copilot pricing for Microsoft 365 has remained steady even as model costs have fallen. Watch whether the MAI model launch triggers a repricing of Copilot for Business — at £25/user/month, it is currently priced above what the underlying model cost would suggest. Increased model competition could force that price down, making AI assistance within Office products more accessible to small teams.