Xiao Hong

AI · AI Agent
肖弘Red XiaoChinaAge ~34

Founder and CEO of Manus / Butterfly Effect · Founder of Nightingale Technology

He registered his first company in his senior year, found footing through WeChat tools after campus products failed, then deliberately moved into global AI, tested distribution with Monica, and pushed self-reported ARR past $100 million with Manus in eight months.

Started while in collegeSerial founderScaled company$100M+ · 2025-12VC-backedChina to globalFrom mainland ChinaBorn 1992Top-school trained

Key numbers

Each figure includes its definition, date, and evidence level.

1992Born · age ~34
~23First venture · 2015
Self-reported$100M+↗Manus ARR · 2025-12
Self-reported$125M+↗total company revenue run rate · 2025-12
Media report$400M–500M↗Manus annualized revenue (media-reported) · 2026-05
Media reportabout $4B↗valuation in a new financing round · 2026-09
Media report$75M↗Benchmark-led financing · 2025-05
Self-reported2M+↗B2B users served by Yiban and Weiban · 2023

Key stages

What he was doing at each age

~21

find real users first

~24

switched to productivity tools after failure

~30

left the WeChat ecosystem to change cycles

~33

Manus ignited the agent category

~33

globalized the organization and legal entity

~33

acquisition announced but not completed

Geographic moves

The city behind each change in direction

  1. Jiangxi
    1992 born in a small town in Jiangxi
  2. Wuhan
    ~21 find real users first~23 started a company straight out of his senior year~24 switched to productivity tools after failure~29 founder and CEO of Nightingale Technology
  3. Beijing
  4. global markets
    ~30 left the WeChat ecosystem to change cycles~31 founder and CEO of Butterfly Effect~33 Manus ignited the agent category~33 founder and CEO
  5. China
  6. Singapore
    ~33 globalized the organization and legal entity
  7. global markets
    ~33 founder and CEO
  8. Manus
  9. Meta system
    ~33 acquisition announced but not completed
  10. Beijing
    ~34 founder and CEO of Manus
  11. Singapore
    ~34 founder and CEO of Manus
  12. China
    ~34 founder and CEO of the independent company~34 founder and CEO~34 founder and CEO

Selected work

a general-purpose AI agent that autonomously decomposes and executes tasks in a cloud computer

Cue

a standalone personal-agent app for phones and desktops, launched with Manus 2.0 in September 2026

a browser AI workspace aggregating multiple models, monetized through individual subscriptions and advanced credits

Yiban Assistant

a WeChat Official Account editing productivity tool, born from a shift in demand after the early campus products failed

Weiban Assistant

a customer-operations tool for the enterprise WeChat environment

How the business works

category
AI · AI agents / productivity tools
users and payers
Three generations of products, three types of payers: Yiban / Weiban are paid for by operations and sales teams (B2B), Monica by individual subscribers, and Manus by individuals and teams paying per task
pricing model
Monica uses Pro / Max / Ultra plans plus advanced credits (an extra 4,000 credits costs $10); from March 2026, Manus Pro starts at $20/month with 4,000 credits, another tier at $40/month with 8,000 credits, and Team at $20/seat/month
second revenue layer
Once a WebDev app is deployed, hosting, databases, AI features, and API calls are billed separately based on actual resource use, preventing long-term hosting costs from being absorbed by a single subscription
main costs
Upstream model APIs, generation services, and channel commissions; Manus credits are mainly consumed by LLM tokens, cloud virtual machines, and third-party APIs. In September 2026, the company said its 2.0 Cascade framework reduced operating costs by 32% (company self-test)
level of validation
scale validatedThe company self-reported ARR above $100 million and total run rate above $125 million in December 2025; it has not disclosed gross margin, failed-task rate, retention, or profit, so this cannot establish that the unit economics work
moat
Solves cold start by acquiring existing plugins; the aggregated experience can create a spread (analytical judgment)
key risks
Upstream vendors may copy features or change prices at any time; having many users does not equal SaaS profitability

Xiao Hong's three generations of products correspond to three pricing logics. Yiban and Weiban are B2B tools embedded in the WeChat ecosystem: operations and sales teams pay for formatting, customer follow-up, and private-domain workflows. Public sources support the company's figure of more than two million B2B users, but there are no current prices, revenue, or gross-margin figures, so having many users cannot be presented directly as SaaS profitability.

Monica is a multi-model aggregation subscription for individuals. With one subscription, users access different models, translation, summarization, image and video generation, and lightweight agents; Pro, Max, and Ultra plans use advanced credits to limit expensive tasks, with an extra 4,000 credits costing $10. Official credit rules It solved cold start by acquiring existing plugins. Its main costs are upstream model APIs, generation services, and channel commissions. The aggregated experience can create a spread, but upstream vendors may copy features or change prices at any time.

Manus brings the billing unit closer to task costs. Official pricing information from March 2026 shows Pro starting at $20/month with 4,000 credits, another tier at $40/month with 8,000 credits, and Team starting at $20/seat/month. Official pricing Credits are mainly consumed by LLM tokens, cloud virtual machines, and third-party APIs; the more complex the task and the longer it runs, the higher the cost. Credit cost rules

Manus also has a second usage-based revenue layer: subscriptions cover “building and execution,” while once users deploy WebDev apps, hosting, databases, AI features, and API calls are billed separately based on actual resource use. WebDev billing This prevents long-term hosting costs from being absorbed by a single subscription. The company self-reported ARR above $100 million and total revenue run rate above $125 million in December 2025, demonstrating scale; but it has not disclosed gross margin, failed-task rate, retention, or profit, so this cannot establish that the unit economics work.

Manus 2.0 went straight after costs in September 2026: the company said its new Cascade framework used 23.2% fewer tokens than the old system, cut task time by 28.2%, and reduced operating costs by 32%. Manus 2.0 announcement These are company self-test figures with no third-party verification; the pricing of the resident cloud computer and standalone Cue app was also not disclosed in the announcement.

Timeline

  1. born in a small town in Jiangxi

    Starting outChina · Jiangxi

    Birth records conflict: sources including Changjiang Daily list 1992, while some list 1993; the county is variously reported as Anyuan or Suichuan. The database currently records 1992 based on the relatively direct source, without specifying the county.

    Source ↗
  2. built campus WeChat products at Huazhong University of Science and Technology

    About 21 years oldStarting outJiangxi → Wuhan · software engineering student · campus product lead

    Built campus weather, WeChat-wall, and similar projects around WeChat Official Accounts, forming his first loop of spotting platform changes, building quickly, and finding real users.

    Source ↗
  3. registered Wuhan Nightingale Technology in his senior year

    About 23 years oldStarting outWuhan · founder and CEO

    He did not take a full-time role at a mature company after graduation; his career began directly as a founder. The exact date comes from an interview with him.

    Source ↗
  4. shifted to Yiban Assistant after the campus products failed and cash nearly ran out

    About 24 years oldTurning pointWuhan · founder and CEO of Nightingale Technology

    The team was once preparing to accept offers from major tech companies, then narrowed the problem from broad campus social networking to Official Account editing efficiency and received its first investment from ZhenFund.

    Source ↗
  5. Yiban and Weiban formed a WeChat-ecosystem product suite

    About 29 years oldMilestoneWuhan · founder and CEO of Nightingale Technology

    A HUST alumni page said that by 2023 the two products had served more than two million B2B users in total; this is a figure provided by the alumni page and the company, with no independent audit found.

    Source ↗
  6. founded Butterfly Effect and moved into global AI applications

    About 30 years oldTurning pointWuhan / Beijing → global markets · founder and CEO

    Deliberately switched from an established domestic SaaS position to AI and overseas markets; when the team was founded, none of the founders had lived abroad for an extended period.

    Source ↗
  7. Monica validated overseas distribution and payment

    About 31 years oldMilestoneglobal markets · founder and CEO of Butterfly Effect

    The product acquired its cold-start users by integrating or acquiring existing browser plugins. An investor recalled that its users grew from thousands to millions and ARR exceeded several million dollars; this is an investor's recollection, not audited financial reporting.

    Source ↗
  8. officially launched the general-purpose AI agent Manus

    About 33 years oldMilestoneglobal markets · founder and CEO of Manus

    The product turned “answering questions” into autonomously decomposing and executing tasks in a cloud computer; its invite-only beta quickly attracted global attention.

    Source ↗
  9. closed a $75 million financing led by Benchmark

    About 33 years oldMilestoneglobal markets · founder and CEO

    Media reported a valuation of about $500 million; the financing amount and valuation are company figures, not Xiao Hong's personal income.

    Source ↗
  10. moved the company's headquarters and key team to Singapore

    About 33 years oldTurning pointChina → Singapore · founder and CEO

    The company closed its China office, shifting from “building a global product in China” to globalizing its organization and legal entity as well; there is no reliable confirmation that Xiao Hong also moved to Singapore long term.

    Source ↗
  11. Manus officially self-reported ARR above $100 million

    About 33 years oldMilestoneSingapore / global markets · founder and CEO

    Eight months after launch, ARR exceeded $100 million, the company's total revenue run rate exceeded $125 million, and the team numbered 105; this is an annualized run rate, not confirmed full-year revenue, and no audit has been found.

    Source ↗
  12. Meta announced the acquisition of Manus

    About 33 years oldTurning pointManus → Meta system · founder and CEO of the acquired company

    Neither side disclosed a price; Reuters, citing people familiar with the matter, said the valuation was about $2–3 billion. Regulators later required the transaction to be withdrawn, so this timeline does not mark it as an exit or describe it as a final exit.

    Source ↗
  13. questioned by the National Development and Reform Commission and restricted from leaving the country

    About 34 years oldFailureChina · Beijing · founder and CEO of Manus

    TechCrunch, citing the Financial Times, reported that Xiao Hong and chief scientist Yichao Ji were questioned by the National Development and Reform Commission that month and told not to leave the country during the review; no formal charges were reported, and Chinese officials described it as a routine regulatory review. In August 2026, the Financial Times said regulators were preparing to lift the restriction and Xiao Hong told employees he planned to return to Singapore, but no official lifting decision had been found and his current travel status remains undisclosed.

    Source ↗
  14. Chinese regulators required the Meta acquisition to be withdrawn

    About 34 years oldFailureChina / Singapore · founder and CEO of Manus

    Cross-border regulation directly changed the outcome; the valuation range reported by the media cannot be used to infer that Xiao Hong personally cashed out.

    Source ↗
  15. Manus officially announced its return to independent operation

    About 34 years oldTurning pointSingapore / China · founder and CEO of the independent company

    On August 11, the company first told users it would “soon return to independence” and arranged to delete some data; it made the formal announcement on September 1, saying the founding team would continue to lead the company. Caixin reported that former investors including Tencent, ZhenFund, and HSG bought the company back for about $2 billion, with Tencent becoming the largest shareholder; the full transaction documents were not public, Xiao Hong's ownership and voting rights were not disclosed, and “continue to lead” refers to management control, not equity control.

    Source ↗
  16. media reported a new financing of about $500 million at a valuation of about $4 billion

    About 34 years oldMilestoneSingapore / China · founder and CEO

    Bloomberg reported that the valuation was about twice the Meta transaction price, with Tencent, HSG, and ZhenFund among the existing shareholders. The company has not announced it, and completion is unconfirmed; the valuation is a company figure, not Xiao Hong's personal income.

    Source ↗
  17. released Manus 2.0 and the personal-agent app Cue

    About 34 years oldMilestoneglobal markets / China · founder and CEO

    The first major release 27 days after returning to independence: the Cascade framework, a resident cloud computer, event-triggered automation, Manus Studio, and the standalone Cue app. The next day, the Global Times reported that the company announced through a public WeChat account that it was forming a team to develop products for the domestic market and work with Chinese model providers.

    Source ↗

Entrepreneurship Was the Career Starting Point

While studying software engineering at Huazhong University of Science and Technology, Xiao Hong built campus weather and WeChat-wall products around public WeChat accounts. On January 20, 2015, while still a senior, he registered Wuhan Nightingale Technology; after graduation he did not take a full-time role at a mature company.

The early campus social and trading products did not grow, cash nearly ran out, and the team at one point prepared to accept offers from large technology companies. The real turning point was not persisting with the original direction but narrowing the problem: he built Yiban around WeChat editing efficiency, then expanded to Weiban for enterprise WeChat. A HUST alumni page says the two products had served more than two million B2B users in total by 2023, but this is not audited data.

Deliberately Replacing an Existing Advantage

In 2022 he founded Butterfly Effect, moving from WeChat-ecosystem SaaS to AI applications for global markets. The team had no ready-made overseas living network, so the choice looked more like deliberately changing the capability boundary than casually internationalizing.

Monica began as a browser AI assistant and solved cold start by integrating or acquiring existing plugins. An investor recalled that the product grew from thousands of users to millions and exceeded several million dollars in ARR. These figures are investor recollections, but at minimum show that before Manus the team had already validated overseas distribution and payment.

Manus's Eight Months

Manus officially launched on March 6, 2025. It changed the interaction from “answering a question” to having an Agent decompose and complete tasks inside a cloud computer, and its invitation-only test quickly attracted global attention. Two months later, Benchmark led a $75 million financing; in the middle of the year, the company moved its headquarters and key team to Singapore.

By December 2025, Manus officially self-reported ARR above $100 million, a total revenue run rate above $125 million, and a team of 105. This speed is a representative result, but the measurement must stay precise: run rate annualizes the revenue pace at that time; it is not confirmed full-year revenue.

An Exit That Did Not Complete

On December 29, 2025, Meta announced the acquisition of Manus, with media reporting a transaction valuation of about $2–3 billion. Chinese regulators subsequently required the transaction to be withdrawn, and on September 1, 2026 Manus officially announced a return to independent operation. Caixin reported that the original investors completed a buyback of about $2 billion, with Tencent becoming the largest shareholder; the complete transaction documents were not public.

During the review, Xiao Hong and chief scientist Yichao Ji were interviewed by the National Development and Reform Commission in March 2026 and told not to leave the country; reports did not show formal charges. In August, reports said the restriction was preparing to be lifted, but there was no official lifting decision, and the independence announcement did not mention the two people's travel status.

This episode therefore cannot be written as “successfully sold to Meta” or “Xiao Hong cashed out billions.” It looks more like an organizational experiment rewritten by geopolitical regulation: the product had globalized and the company had moved to Singapore, but the founder's control and exit path remained constrained by cross-border policy.

2026: Manus 2.0 and Cue, a Bet on the Resident Agent

On September 28, 2026, Manus released Manus 2.0, only 27 days after returning to independence. The company described it as a new architecture rather than a version update: its in-house Cascade framework handles orchestration, the cloud computer keeps projects running in a dedicated environment, automation is triggered by external events, and Manus Studio puts video editing and game development into a desktop workbench.

It also launched Cue, a standalone personal-agent application for phones and desktops. Both point in the same direction: an Agent is no longer merely “give it one task and wait for completion,” but stays online and handles matters proactively for the user. “Resident Agent” is this site's summary, not official terminology.

Another line is a return to the Chinese market. On the day after the release, the company announced through a public WeChat account that it was forming a team for domestic-market products and would work with Chinese model providers. It had closed its China office more than a year earlier; Tencent is now the largest shareholder, and Bloomberg reported a valuation of about $4 billion for the new financing, but the company has not yet announced it officially.

In their words
I registered my first company on January 20, 2015, when I was still a senior.
—— Xiao Hong ↗

More quotes

Chinese entrepreneurs today should globalize more aggressively.
#global marketsSource ↗
Joining Meta allows us to build on a stronger, more sustainable foundation.
2025#acquisitionSource ↗

wording from Meta's acquisition announcement; the transaction was later required to be withdrawn.

My read

Xiao Hong is easiest to write as a model founder who suddenly built Manus. In reality, his core capability is closer to product and distribution: the campus WeChat projects, Yiban, Weiban, and Monica all found demand quickly as platforms changed, acquired an initial user base, and moved existing distribution into the next product. Manus's breakout continued a decade of product trial and error rather than appearing from nowhere.

The first thing worth copying is low-cost, continuous validation of real use: after the campus social product failed, he did not defend a grand narrative, but narrowed the problem to WeChat editing efficiency. Second, after gaining a foothold in WeChat SaaS, he still deliberately switched into AI and global markets. Third, Monica solved cold start by integrating existing plugins, then moved users, payments, and team capabilities into a larger Agent product.

The business mechanism is consistent: Yiban and Weiban sit inside the WeChat workflow, Monica aggregates several models into one subscription, and Manus upgrades answers into cloud task execution. Manus uses credits to map the variable costs of LLMs, virtual machines, and third-party APIs to user allowances; WebDev deployment is charged separately for cloud resources. This is closer to real costs than a single unlimited subscription.

There are three cautions. First, this is an application business highly dependent on upstream models and platform rules rather than a core-model route. Second, $100 million ARR and a $125 million revenue run rate are company-reported annualized speeds; the company has not disclosed gross margin, retention, or profit. Third, the Meta transaction's announcement and withdrawal show that cross-border regulation can rewrite ownership, and the founder himself was restricted from leaving the country during review. The ending has another reversal: the company that went overseas most thoroughly was ultimately bought back by Chinese capital and began rebuilding a domestic team; the $4 billion valuation reported by media has not been officially announced and is not the founder's personal result. The birth year also has a 1992/1993 conflict; this site records 1992 for now and does not treat derived age as precise fact.

Next founder