Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, knowledge, and safety leaders. Subscribe Now
Singaporean AI startup Manus, which made headlines earlier this yr for its strategy to a multi-agent orchestration platform for shoppers and “professional”-sumers (professionals eager to run work operations), is again with an fascinating new use of its know-how.
Whereas many different main rival AI suppliers akin to OpenAI, Google, and xAI which have launched “Deep Analysis” or “Deep Researcher” AI brokers that conduct minutes or hours of intensive, in-depth net analysis and write well-cited, thorough experiences on behalf of customers, Manus is taking a special strategy.
The firm simply introduced “Huge Analysis,” a brand new experimental characteristic that allows customers to execute large-scale, high-volume duties by leveraging the ability of parallelized AI brokers — much more than 100 at a single time, all centered on finishing a single process (or sequence of sub-tasks laddering up stated overarching purpose).
Manus was beforehand reported to be utilizing Anthropic Claude fashions to energy its platform.
The AI Affect Collection Returns to San Francisco – August 5
The subsequent section of AI is right here – are you prepared? Be part of leaders from Block, GSK, and SAP for an unique take a look at how autonomous brokers are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.
Safe your spot now – area is restricted: https://bit.ly/3GuuPLF
Parallel processing for analysis, summarization and artistic output
In a video posted on the official X account, Manus co-founder and Chief Scientist Yichao ‘Peak’ Ji reveals a demo of utilizing Huge Analysis to check 100 sneakers.
To finish the duty, Manus Huge Analysis almost immediately spins up 100 concurrent subagents — every assigned to investigate one shoe’s design, pricing, and availability.
The result’s a sortable matrix delivered in each spreadsheet and webpage codecs inside minutes.
The corporate suggests Huge Analysis isn’t restricted to knowledge evaluation. It will also be used for inventive duties like design exploration.
In a single situation, Manus brokers concurrently generated poster designs throughout 50 distinct visible types, returning polished property in a downloadable ZIP file.
In accordance with Manus, this flexibility stems from the system-level strategy to parallel processing and agent-to-agent communication.
Within the video, Peak explains that Huge Analysis is the primary utility of an optimized virtualization and agent structure able to scaling compute energy 100 occasions past preliminary choices.
The characteristic is designed to activate mechanically throughout duties that require wide-scale evaluation, with no guide toggles or configurations required.
Availability and pricing
Huge Analysis is offered beginning right this moment for customers on Manus Professional plan and can steadily develop into accessible to these on the Plus and Primary plans. As of now, subscription pricing for Manus is structured as follows per 30 days.
- Free – $0/month Consists of 300 each day refresh credit, entry to Chat mode, 1 concurrent process, and 1 scheduled process.
- Primary – $19/month Provides 1,900 month-to-month credit (+1,900 bonus throughout restricted provide), 2 concurrent and a pair of scheduled duties, entry to superior fashions in Agent mode, picture/video/slides era, and unique knowledge sources.
- Plus – $39/month Will increase to three concurrent and three scheduled duties, 3,900 month-to-month credit (+3,900 bonus), and consists of all Primary options.
- Professional – $199/month Gives 10 concurrent and 10 scheduled duties, 19,900 credit (+19,900 bonus), early entry to beta options, a Manus T-shirt, and the total characteristic set together with superior agent instruments and content material era.
There’s additionally a 17% low cost on these costs for customers who want to pay up-front yearly.
The launch builds on the infrastructure launched with Manus earlier this yr, which the corporate describes as not simply an AI agent, however a private cloud computing platform.
Every Manus session runs on a devoted digital machine, giving customers entry to orchestrated cloud compute via pure language — a setup the corporate sees as key to enabling true general-purpose AI workflows.
With Huge Analysis, Manus customers can delegate analysis or inventive exploration throughout dozens and even a whole lot of subagents.
Not like conventional multi-agent methods with predefined roles (akin to supervisor, coder, or designer), every subagent inside Huge Analysis is a totally succesful, absolutely featured Manus occasion — not a specialised one for a particular function — working independently and capable of tackle any normal process.
This architectural choice, the corporate says, opens the door to versatile, scalable process dealing with unconstrained by inflexible templates.
What are the advantages of Huge over Deep Analysis?
The implication appears to be that working all these brokers in parallel is quicker and can end in a greater and extra various set of labor merchandise past analysis experiences, versus the only “Deep Analysis” brokers different AI suppliers have proven or fielded.
However whereas Manus promotes Huge Analysis as a breakthrough in agent parallelism, the corporate doesn’t present direct proof that spawning dozens or a whole lot of subagents is more practical than having a single, high-capacity agent deal with duties sequentially.
The discharge doesn’t embody efficiency benchmarks, comparisons, or technical explanations to justify the trade-offs of this strategy — akin to elevated useful resource utilization, coordination complexity, or potential inefficiencies. It additionally lacks particulars on how subagents collaborate, how outcomes are merged, or whether or not the system gives measurable benefits in pace, accuracy, or value.
Consequently, whereas the characteristic showcases architectural ambition, its sensible advantages over less complicated strategies stay unproven primarily based on the knowledge supplied.
Sub-agents have a blended observe file extra usually, up to now…
Whereas Manus’s implementation of Huge Analysis is positioned as an development basically AI agent methods, the broader ecosystem has seen blended outcomes with related subagent approaches.
For instance, on Reddit, self-described customers of Claude’s Code have raised considerations about its subagents being sluggish, consuming giant volumes of tokens, and providing restricted visibility into execution.
Frequent ache factors embody lack of coordination protocols between brokers, difficulties in debugging, and erratic efficiency throughout high-load durations.
These challenges don’t essentially replicate on Manus’s implementation, however they spotlight the complexity of creating strong multi-agent frameworks.
Manus acknowledges that Huge Analysis remains to be experimental and should include some limitations as improvement continues.
Wanting forward
With the rollout of Huge Analysis, Manus deepens its dedication to redefining how customers work together with AI brokers at scale.
As different platforms wrestle with the technical challenges of subagent coordination and reliability, Manus’s strategy might function a take a look at case for whether or not generalized agent cases — slightly than narrowly scoped modules — can ship on the imaginative and prescient of seamless, multi-threaded AI collaboration.
The corporate hints at broader ambitions, suggesting that the infrastructure behind Huge Analysis lays the groundwork for future choices. Customers and business watchers alike will probably be paying shut consideration as to whether this new wave of agent structure can reside as much as its potential — or whether or not the challenges seen elsewhere within the AI area will ultimately catch up.
Correction: This text initially incorrectly said that Manus was primarily based in China when it’s not; it’s in Singapore. Additionally it cited prior reporting that it used Alibaba Qwen fashions; it doesn’t. We’ve got up to date and remorse the errors.
