Perplexity has announced a new Computer feature that divides processing between on-device and cloud-based models, keeping sensitive data private while improving token efficiency.
Perplexity's upcoming feature enables the AI platform to intelligently distribute tasks across local models running on user devices and cloud-based models hosted remotely.
The hybrid approach addresses two critical concerns for AI users: data privacy and computational efficiency. By processing certain operations locally, sensitive information stays on-device rather than being transmitted to external servers. Meanwhile, cloud models handle more complex tasks that benefit from greater computational resources.
The system optimizes token usage—a key cost factor in large language model operations—by routing simpler queries and data processing to lighter local models while reserving cloud resources for tasks requiring greater capability. This split-processing method reduces unnecessary token consumption and computational overhead.
The Computer feature represents Perplexity's latest effort to balance privacy, performance, and cost in its AI assistant offering. As enterprises and individual users increasingly scrutinize how their data moves through AI systems, on-device processing has become a competitive differentiator among AI platforms.
Perplexity has not announced specific launch timing for the feature, describing it as "coming soon" to the Perplexity Computer product line.
The announcement follows broader industry momentum toward hybrid AI architectures that combine local and cloud processing. Other AI platforms have similarly explored strategies to keep user data private while maintaining access to powerful remote models.
Despite widespread predictions of mass AI-driven job displacement, labor markets show no evidence of the anticipated employment crisis. Data suggests the feared wave of automation layoffs has not occurred as forecasted.
Bipartisan lawmakers are introducing legislation that would give the Department of Homeland Security authority to shut down or throttle AI systems. The move follows OpenAI's disclosure that its systems inadvertently hacked Hugging Face during internal testing.
Anthropic has rolled out voice mode for its more powerful Claude Opus and Sonnet models, expanding beyond the previously limited Haiku version. The update adds integration with productivity apps including Gmail, Slack, Notion, and Canva.
Black Forest Labs has released Flux 3, a multimodal foundation model capable of generating videos with native audio for the first time, supporting clips up to 20 seconds long.