Breaking

Perplexity Announces Mac Hybrid Mode for Computer, Using Local PPLX Qwen 3.8 27B

September 1, 2026 at 11:32 EDT

AI Agents
Foundation Models

Perplexity has announced Hybrid mode for Perplexity Computer on Mac, a local-cloud setup that delegates selected work to on-device inference while retaining cloud orchestration. The feature also adds Privacy Gate, which checks for personally identifiable information (PII) before data is sent to the cloud.

AUG 31, 2026 · PERPLEXITY

Hybrid mode brings local subtasks to Mac

Perplexity Computer keeps orchestration in the cloud while handing suitable work to a local model. Privacy Gate checks outbound data for PII before transmission.

Local model footprint

Mac options range from about 5.6 GB to 19 GB. Each block represents approximately 1 GB.

5.6 GB

Gemma-based

17.4 GB

Qwen 32B

19 GB

Perplexity option

What the hybrid design changes

27B

PPLX Qwen 3.8 scale

16 GB

Smallest Mac memory target

32 GB

Larger-model memory target

85.4%

PPLX 27B on a 53-task local benchmark

How a task moves

Cloud orchestrator →

Local model handles suitable work →

Privacy Gate checks outbound data →

User decides on cloud transfer

Upside

Suitable local work can avoid Computer credits, while PII screening adds a decision point before cloud transmission.

Trade-offs

Cloud orchestration remains part of the design, harder reasoning may still need a frontier model, and flagged transfers require user judgment.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.