Microsoft's New AI Assistant Runs on Chinese Qwen3.5-9B Model

What happened

Microsoft has confirmed that its newest in-house AI system, internally referred to as Microsoft-Decision-1, is built on top of Qwen3.5-9B — an open-weight model released by a Chinese AI lab — rather than on proprietary foundation-model technology developed by Microsoft itself. According to reporting by The Register, the system is designed to compete with a rival AI assistant referred to in coverage as "Jev."

Microsoft has indicated that this first release is a transitional step: the company says a subsequent version of Decision-1 will be rebuilt on homegrown model technology, rather than continuing to rely on an externally sourced open-weight base.

Why it matters

The move illustrates how deeply open-weight models — including those released by Chinese labs — have become embedded in the toolkits of even the largest Western technology vendors. Rather than waiting for a fully proprietary model to mature, Microsoft has opted to ship a competitive product now by fine-tuning an available open-weight base, then signalling a roadmap toward owned technology later. That is a notable shift in how "build versus adopt" decisions are being made at the frontier of enterprise AI.

For organisations evaluating AI vendors, it is a reminder that the underlying model powering a branded assistant or copilot is not always developed solely by the company whose name is on the product. Procurement, security and governance teams assessing AI tools may increasingly need to ask not just "what can this model do," but "whose weights is it actually running on," since that affects data handling, model provenance and long-term roadmap stability.

By the numbers

• 9 billion parameters: the scale of the Qwen3.5-9B open-weight model underpinning the first version of Microsoft-Decision-1.

The Renascence take

The headline detail — a Chinese open-weight model inside a flagship Microsoft AI product — will draw most of the attention. The more useful signal for operators is what it says about sequencing: ship a credible experience now on borrowed foundations, then migrate the engine underneath once proprietary technology catches up, all while keeping the front-end brand and user experience stable.

Most customers will never know, or care, which model sits behind an assistant — what they judge is whether it is fast, accurate and trustworthy today. The lesson for experience leaders is to decouple the roadmap conversation from the delivery conversation: it is entirely viable to launch on a capable third-party foundation while being transparent internally about a planned transition, provided governance, data-handling and quality bars are met throughout. The real discipline is making sure that switch, when it comes, is invisible to the end user.