Hybrid Intelligence
A hybrid intelligence architecture introduced by Microsoft for Windows Copilot. Its core design enables cloud-based large language models and on-device models to work collaboratively, dynamically selecting the compute location based on task characteristics to balance capability, cost, privacy, and latency.
Timeline (last 90 days)
Microsoft announced the introduction of a new capability called Hybrid Intelligence for Copilot on the Windows platform
The core idea of Hybrid Intelligence is to enable cloud-based large models and local models to work collaboratively, dynamically selecting the computation location based on the nature of the task
Hybrid Intelligence encompasses three layers of capability: reading PC context, performing operations on behalf of the user, and coordinating between local models and the cloud
The information officially released so far is relatively general, and details such as which tasks local models can handle, cloud scheduling logic, NPU hardware requirements, and licensing granularity have not yet been clarified.
All Facts (4)
The core idea of Hybrid Intelligence is to enable cloud-based large models and local models to work collaboratively, dynamically selecting the computation location based on the nature of the task
50%UnverifiedHybrid Intelligence encompasses three layers of capability: reading PC context, performing operations on behalf of the user, and coordinating between local models and the cloud
50%UnverifiedThe information officially released so far is relatively general, and details such as which tasks local models can handle, cloud scheduling logic, NPU hardware requirements, and licensing granularity have not yet been clarified.
50%UnverifiedMicrosoft announced the introduction of a new capability called Hybrid Intelligence for Copilot on the Windows platform
50%