Microsoft is considering Moonshot AI’s Kimi K3 model for some Copilot workloads. An internal estimate suggests it could save up to $600 million in computing costs.
The company is also working to make Kimi K3 available through its Azure cloud platform. This is an evaluation and access-preparation effort, not an announced migration of Copilot to the model.
Microsoft’s interest is tied to inference costs — the computing required to process user requests. A cheaper model could be particularly relevant to Copilot Cowork, where pricing depends on the number of tokens actually used.
Kimi K3 is a multimodal open-weight model. Available technical descriptions list 2.8 trillion parameters and a context window of about 1 million tokens.
The model is being considered as a possible supplement to OpenAI and Anthropic systems for selected use cases. There is no confirmation that Microsoft will abandon its current providers or has formed a partnership with Moonshot AI.
