Manus Launchs Wide Research Parallel AI at Scale

Manus Unveils Parallel AI with Wide Research

Chinese AI startup Manus has launched Wide Research, an experimental feature that leverages more than 100 concurrent AI subagents to execute large-scale analyses and creative tasks in parallel. This system scales compute power up to 100x beyond typical single-agent approaches, delivering structured results in minutes.

Parallel Processing at Scale

Unlike Deep Research offerings that run one high-capacity agent sequentially, Wide Research spins up dozens or hundreds of fully featured Manus instances simultaneously. Each subagent tackles a sub-task — from product data scraping to design exploration — then communicates findings back into a unified matrix or asset bundle.

• Rapid comparison of products, prices, and availability
• Mass generation of creative assets across diverse styles
• Automated extraction and formatting into spreadsheets or web layouts

Real-World Applications

In a demo, Manus co-founder Yichao “Peak” Ji showed how Wide Research compared 100 sneaker models by analyzing design, pricing, and stock availability concurrently, returning a sortable matrix in minutes. In another scenario, 50 distinct poster designs were generated in parallel, delivered as a ZIP file.

Pricing Tiers and Access

Wide Research is live for Pro subscribers and will roll out to Plus and Basic plans. Manus pricing starts with a Free tier at $0/month, Basic at $19, Plus at $39, and Pro at $199, each offering increasing concurrent tasks, credits and early beta access.

Assessing the Trade-Offs

While Wide Research promises speed and output variety, Manus hasn’t released benchmarks comparing parallel subagents to single-agent processing. Questions remain around resource consumption, coordination complexity, and merging results. The ecosystem’s mixed track record on subagent frameworks suggests careful validation is essential.

Looking Ahead

Manus positions Wide Research as the foundation for future general-purpose AI workflows, built on a personal cloud computing platform with virtual machines for each session. As enterprises explore multi-agent orchestration, the effectiveness of generalized subagents versus specialized roles will be a critical measure of success.