Microsoft reportedly considers Kimi K3 for Copilot to cut costs

Microsoft is reportedly testing Moonshot AI's Kimi K3 model internally as it considers incorporating it into its Copilot service. The tech giant is assessing if a portion of Copilot's inference requests, currently handled by models from OpenAI and Anthropic, could be moved to Kimi K3. The Information first reported on the evaluation.

Estimates suggest that shifting some of Copilot's workload to Kimi K3 could lead to annual cloud infrastructure cost savings of up to $600 million. This move aligns with Microsoft's broader strategy to diversify its AI model usage, reduce reliance on single suppliers, and manage the escalating costs associated with serving high volumes of AI requests.

The Kimi K3 model is being assessed for its performance in areas like coding and reasoning, as well as its cost-efficiency for specific Copilot tasks. Any potential deployment would still require addressing technical integration, data sovereignty, security, and export control considerations, and might initially be limited to less sensitive applications.

Microsoft has not yet made a final decision regarding the replacement of any existing models.