Tuesday, 1 September 2026
Rīga TV

World and Latvian news in one place

TechnologyPublished: 1 September 2026 at 18:33

Perplexity launches Hybrid Compute to split tasks between cloud and local AI

Perplexity has introduced Hybrid Compute, a tool that processes sensitive data locally on a user's computer while offloading the rest of a task to more powerful cloud AI models. The feature is currently limited to Apple Silicon Macs.

Foto: Engadget

Perplexity has unveiled a new feature called Hybrid Compute, expanding on its AI agent platform that began in February with the launch of Perplexity Computer. That system, similar in concept to Claude Cowork, allows AI agents to autonomously carry out tasks using the web as well as files and apps stored on a user's computer.

Hybrid Compute lets users split a single task between a frontier cloud-based model, such as Opus 5 or GPT-5.6 Sol, and a local language model running directly on their machine. The goal is to keep any sensitive information processed by the local model, so it never leaves the device, while the cloud model handles the rest of the workload.

Use cases

Perplexity points to an example of a lawyer preparing a legal brief by comparing a case against existing case law, where client data could remain confidential by staying local. The company also frames the feature as a cost-saving option, since offloading part of a task away from pricier frontier models can reduce inference expenses.

According to Jon Staff, who oversees Perplexity's Mac products, the feature is built directly into the Mac app and automatically scans uploaded files or shared information for sensitive content, prompting users before anything is sent to the cloud. To support this, Perplexity trained a new privacy classifier that suggests which files or data should stay on the local machine. Users can review and adjust these suggestions, and also choose which models handle which part of the work.

For local processing, available options include Gemma E4B and two versions of Qwen's 35-billion-parameter 3.6 model, one of which was further trained by Perplexity itself. Installing a local model requires no terminal use, as the app manages installation automatically. While a task runs, users can view CPU, GPU and memory usage, along with a token counter — and local model output does not incur any charges.

Staff acknowledged that a fully cloud-based approach generally produces better results, but said some users prioritize privacy and lower costs over maximum capability, and that Perplexity aims to give users control over that tradeoff.

Hybrid Compute is currently available only on Apple Silicon Macs running macOS 15, with at least 32GB of unified memory recommended. It is offered to Pro and Max subscribers as well as enterprise customers.

Comments

0/1500

Comments are automatically moderated. No hate, threats, personal data or spam.

Loading comments…

More in this category