Perplexity Computer adds local Nemotron routing to cut cloud token spend

Read the original article →

What happened

Perplexity updated Perplexity Computer on July 20 with automatic routing of eligible tasks to NVIDIA Nemotron 3 models running locally on Copilot+ PCs. The system falls back to cloud frontier models only when needed.

CEO Aravind Srinivas said the change cut average session cost 45 percent for beta users while keeping quality stable on research and drafting workflows. The feature requires Windows Copilot+ hardware with an NPU and pairs with Perplexity's June split-local-cloud architecture. Mac support is planned for Q4.

Why it matters

Hybrid local-cloud routing is emerging as a practical response to frontier pricing and gating. Nemotron on-device gives users a no-API-cost tier for routine work while preserving cloud access for harder tasks.

Hardware requirements limit adoption today, but the pattern may spread to other agent desktops.

MintedBrain take

Cost savings depend on how often your workflows trigger cloud fallback. Track fallback rates for a week before assuming the 45 percent figure applies to your team, especially for coding or data-heavy tasks.

References

Discussion

  • Loading…

← Back to News