Manus, the autonomous AI agent platform that made headlines with its March 2025 debut, has unveiled its most significant update yet with the launch of 'Wide Research' - a revolutionary approach to AI-powered research and data processing.
While competitors like OpenAI, Google, and xAI have focused on 'Deep Research' capabilities that use a single AI agent to conduct extensive, sequential analysis, Manus has taken a fundamentally different approach. Wide Research enables users to deploy swarms of over 100 AI agents simultaneously, each working on different aspects of a complex task.
What sets Wide Research apart is its architecture. Unlike traditional multi-agent systems with predefined specialized roles, every subagent in Wide Research is a fully capable, general-purpose Manus instance operating independently on its own virtual machine. This flexibility allows the system to adapt to diverse domains without being constrained by rigid formats.
In a demonstration by Manus co-founder and Chief Scientist Yichao 'Peak' Ji, the system compared 100 different sneakers by instantly deploying 100 concurrent subagents - each analyzing one shoe's design, pricing, and availability. The result was a sortable matrix delivered in both spreadsheet and webpage formats within minutes. Another showcase generated 50 unique poster designs simultaneously, returning them as a downloadable ZIP file.
"Wide Research marks a milestone in our exploration of the scaling laws of AI agents, but it's just the beginning," stated Manus in their announcement. The infrastructure behind this feature is part of the company's broader vision to democratize access to cloud computing power through natural language interactions.
The feature is currently available to users on the Manus Pro plan ($199/month) and will gradually roll out to Plus ($39/month) and Basic ($19/month) tiers. While the technology shows immense promise for applications in finance, academia, legal research, and creative fields, Manus acknowledges that Wide Research is still experimental, with some limitations in its current implementation.