Perplexity has introduced a new hybrid inference system that automatically decides whether to run artificial intelligence tasks on a user's own device or send them to the cloud. The company says the approach is designed to give users better privacy and lower costs while also cutting its own server bills.
What the system does
The system routes each AI request based on factors like complexity and sensitivity. Simple or private queries stay on the device, while heavier tasks go to the cloud. Perplexity hasn't detailed the exact logic that triggers a switch, but the goal is to keep as much processing local as possible.
Privacy and cost benefits
Keeping sensitive data on the device reduces the amount of information that travels to remote servers. That matters for users who don't want their conversations or documents stored or analyzed off-device. At the same time, offloading simple tasks to the device lowers Perplexity's cloud costs — a win for the company's bottom line.
For users, the hybrid model could mean fewer subscription fees or less bandwidth consumption, though Perplexity hasn't announced new pricing tiers. The company says the system is already running in production, so current users may see changes in response times or data usage.
How it's different
Most AI services today process everything in the cloud, which can be expensive and slow for simple requests. A few others run entirely on-device but can't handle big models. Perplexity's hybrid sits in the middle, splitting the work automatically.
That flexibility also means the company can update or swap out the on-device model without requiring a full app download, as long as the cloud handles the heavy lifting for new features.
Unresolved questions
Perplexity hasn't said which specific tasks will stay local or how much privacy the on-device processing actually provides — the local model still needs to be downloaded and could potentially be inspected. The company also hasn't released benchmarks showing real-world cost savings for users or its own infrastructure.
For now, the system is live. How much it changes the user experience will depend on how aggressively Perplexity pushes tasks to the device.
