Perplexity introduces hybrid compute for enhanced data privacy and cost control

Perplexity has launched Hybrid Compute, a new feature that enables users to split tasks between cloud-based models and local large language models. This allows sensitive data to remain on the user's device, enhancing privacy. Jon Staff, Product Manager of Perplexity Mac, explained that the system automatically checks for sensitive content before uploading files or sending messages. Perplexity has also trained a privacy classifier to suggest files for local storage. Users can review and confirm the files before assigning tasks to different models. Currently available local models include Gemma E4B and two versions of Qwen 3.6, both with 35 billion parameters. Perplex, Staff noted, that while cloud models produce better results, privacy and cost may be more important for some users. Hybrid Compute is available for Macs with Apple Silicon chips running macOS 15, requiring at least 32GB of unified memory. It is accessible to Pro and Max subscribers and enterprise customers.