Butterfly Effect

Updated June 2026

Butterfly Effect is the Singapore-headquartered company behind Manus, the autonomous AI agent that went viral in 2025. Founded in China by Xiao Hong, it builds no foundation models of its own — it orchestrates third-party models such as Claude and Qwen. In December 2025 Meta agreed to acquire it for more than $2 billion, but China's regulator blocked the deal and Meta began unwinding it in June 2026.

Website →API →Pricing →

Butterfly Effect is the company behind Manus, the autonomous AI agent that became 2025’s breakout product. Founded in China in 2022 by Xiao Hong and now headquartered in Singapore, Butterfly Effect is unusual among AI companies in one defining way: it builds no foundation models of its own. Manus is an orchestration layer that routes tasks to third-party frontier models — principally Anthropic’s Claude and fine-tuned versions of Alibaba’s Qwen — and turns them into an agent that completes multi-step work.

That model-agnostic design made Manus possible and, in 2026, made it a geopolitical flashpoint. In December 2025 Meta agreed to buy Butterfly Effect for a reported $2 billion-plus, adding millions of paying users in one of the year’s most-watched AI acquisitions (Wall Street Journal). Four months later China’s regulator blocked the deal on national-security grounds, and by June 2026 Meta had begun unwinding it — severing data access and “sunsetting” the integration while the founders explored buying the company back (Bloomberg, CNBC). As of 18 June 2026, ownership is unresolved — but the Manus product keeps operating.

Quick facts

History and founding

Xiao Hong founded Butterfly Effect in 2022, two months before ChatGPT launched (Wikipedia). The company’s first hit was Monica, a 2023 browser extension that put multiple commercial large language models behind one interface for translation, summarisation and writing — an early signal of the “aggregate the best models” philosophy that would define Manus.

The Manus project began in October 2024, drawing inspiration from the San Francisco coding tool Cursor. Its name comes from the Massachusetts Institute of Technology motto “Mens et Manus” — “mind and hand” (Wikipedia). Because most of Butterfly Effect’s engineers were in China and Manus depended on American models unavailable there, the product was designed from the start for markets outside China — chiefly North America, Japan and South Korea.

Manus launched in invitation-only beta on 6 March 2025. Its demo video — showing the agent autonomously screening resumes and analysing stocks — drew more than a million views in twenty hours, and demand for invite codes spawned a resale market where codes reportedly changed hands for between $1,000 and $13,800 (Wikipedia). It was the first time a general-purpose AI agent, rather than a chatbot, became a consumer phenomenon.

Funding and partnerships

Butterfly Effect raised modestly by frontier-lab standards, reflecting that it spends on orchestration and inference rather than training runs.

Total disclosed funding is roughly $85 million through the Series B (Wikipedia). Along the way the founders turned down money more than once: ByteDance reportedly offered to acquire Butterfly Effect for about $30 million in 2024, and several Chinese local governments offered to invest in 2025 — both declined, the latter over concerns that domestic government ties would complicate the company’s Western ambitions.

Butterfly Effect’s most important relationships are not investors but model providers. Manus relies on Anthropic’s Claude for core reasoning and on fine-tuned Alibaba Qwen models for specialised tasks; a planned Chinese version of Manus, announced as a partnership with the Qwen team, was shelved when the company pivoted away from mainland China. Its single largest corporate relationship — the Meta acquisition — is covered below.

Product timeline

Leadership

• Xiao Hong — founder and CEO. Founded Butterfly Effect in 2022 and previously built the Monica browser extension.
• Ji Yichao (“Peak”) — co-founder and chief scientist on Manus. Born in 1992 and raised in Colorado and Beijing, he founded Peak Labs in 2012 with backing from ZhenFund and Sequoia China, built the Mammoth Browser as a teenager, and was named to Forbes China’s 30 Under 30 in 2012 and 2013 (MIT Technology Review).
• Zhang Tao — product director, part of the founding Manus team.

In March 2026, amid the Chinese regulatory review of the Meta deal, the Financial Times reported that authorities had barred Xiao Hong and Ji Yichao from leaving the country under exit-ban orders (Wikipedia).

Business and financials

Butterfly Effect monetises Manus through a credit-based subscription rather than per-seat pricing: a flat fee buys a pool of credits that each agent task consumes (see the Manus app page for the full pricing breakdown). Reported revenue grew quickly — the company’s run rate rose from roughly $90 million in August 2025 to about $125 million by December 2025 (Sacra). When Meta announced the acquisition, it described the deal as adding “millions of paying users” (Wall Street Journal).

The company is lean: after relocating to Singapore in mid-2025, around 40 core technical staff moved with it and most of the roughly 120 China-based employees were laid off, leaving about 100 employees by late 2025 (Wikipedia). Usage skews international — manus.im drew about 13.9 million visits in September 2025, with Brazil its single largest market ahead of the United States and Japan. All revenue and user figures here are press- or analyst-sourced; Butterfly Effect is private and publishes no audited financials.

Competition and market position

Butterfly Effect helped create the consumer “general AI agent” category, and now competes on two fronts. On one side are the big assistants that have bolted agent features on: OpenAI’s ChatGPT (Agent Mode and Operator) and Anthropic’s Claude (the Cowork desktop agent). On the other are dedicated agent start-ups such as Genspark and coding-agent specialists like Cognition’s Devin.

Its position is distinctive. Because Manus orchestrates rather than trains, Butterfly Effect can always reach for the strongest available model and is not tied to one lab’s roadmap — a genuine advantage in a fast-moving field. The flip side is that its capability ceiling is set by the models it rents, and a “GPT wrapper” critique has followed it since launch (MIT Technology Review). Its real moat is product: the agent harness, the cloud execution environment, Wide Research’s parallel sub-agents, and the breadth of output — research, slides, spreadsheets, websites and, since version 1.6, mobile apps.

Controversies

• The “GPT wrapper” debate. Because Manus builds no model of its own, critics questioned whether it was a genuine technical advance or a polished orchestration of others’ models. Supporters counter that the agent harness and execution environment are the hard part — and that the product’s results speak for themselves.
• US Treasury review (May 2025). The Department of the Treasury examined Benchmark’s investment under the Outbound Investment Security Program, which governs US investment in sensitive Chinese technology. The review largely faded after Butterfly Effect moved its headquarters to Singapore and on the basis that it trains no foundation models (Wikipedia).
• Exit bans on the founders (March 2026). During China’s review of the Meta deal, authorities barred Xiao Hong and Ji Yichao from leaving the country.
• China blocks the Meta acquisition (April 2026). China’s National Development and Reform Commission ordered the parties to withdraw the transaction on national-security grounds (BBC, TechCrunch). The case became a marker of how far Washington and Beijing have drifted apart on AI (Fortune).
• Data-residency and ownership uncertainty. Tasks run on Butterfly Effect’s cloud, the company’s China origins draw scrutiny, and the unresolved ownership question makes the controlling entity itself uncertain — all reasons to weigh carefully for sensitive or regulated work.

Recent developments (2026)

The Meta acquisition and its collapse are the company’s defining 2026 story.

• December 2025 — Meta agrees to buy Butterfly Effect for a reported $2 billion-plus, planning to keep operating Manus and fold its technology into Meta AI (Reuters).
• January–March 2026 — China opens a review. The Ministry of Commerce launched an evaluative investigation; by March, exit bans were placed on the two founders.
• 27 April 2026 — China blocks the deal, directing the parties to unwind it entirely.
• June 2026 — Meta begins the unwind. Meta severed Manus’s access to its internal systems, halted data sharing, and an internal memo described “sunsetting” Manus and migrating its work onto Meta’s own systems (Bloomberg, CNBC).
• Founder buyback talks. The three founders are reportedly seeking to raise around $1 billion to buy Manus back from Meta at the same valuation Meta paid. The outcome is unresolved as of 18 June 2026.

Through all of it, the Manus product has continued to operate and ship — Manus 1.6 remains the current release.

Where Butterfly Effect excels

• Category creation. Manus was the first general-purpose AI agent to break through to a mainstream audience.
• Model-agnostic by design. Orchestrating third-party models lets it adopt the best available model per task without committing to one lab.
• Product breadth. Few rivals span autonomous research, full-stack web and mobile app building, slides, spreadsheets and image editing in a single agent.
• Fast iteration. The jump from Manus 1.5 to 1.6 in roughly two months added a stronger flagship agent, mobile development and visual editing.

Where Butterfly Effect falls short

• No model of its own. Its capability ceiling — and its exposure to any access disruption — is set by the third-party models it rents.
• Ownership and geopolitics. The blocked-and-unwinding Meta deal leaves the company’s owner uncertain and puts it squarely in the US-China AI crossfire.
• Pricing opacity. The credit system is powerful but hard to budget, and is the product’s most common complaint.
• Data-residency questions. Cloud execution plus China origins make it a harder sell where strict data governance is mandatory.
• Retention after the hype. Traffic cooled from its March 2025 spike, the test of whether an agent becomes a durable habit rather than a viral moment.

Developer resources

Butterfly Effect offers a developer API for building on the Manus agent, alongside product documentation, a help centre and a trust centre. Manus is also available as desktop apps (macOS, Windows) and mobile apps (iOS, Android), with integrations including Slack and Mail. Pricing and credit details live on the pricing page.

Frequently asked questions

Who owns Manus?

Manus is owned by Butterfly Effect, a Singapore-headquartered company founded in China by Xiao Hong. Meta agreed to acquire Butterfly Effect in December 2025 for a reported $2 billion-plus, but China’s regulator blocked the deal in April 2026 and Meta began unwinding it in June 2026, so ownership is unresolved as of mid-2026. The product continues to operate normally.

Where is Butterfly Effect based?

Butterfly Effect is headquartered in Singapore. It was founded in China and ran offices in Beijing and Wuhan, but relocated its headquarters and core team to Singapore in mid-2025 and cut its mainland-China operations.

Who founded Manus?

Butterfly Effect was founded by Xiao Hong in 2022. Ji Yichao, known as “Peak”, is the co-founder and chief scientist behind the Manus product, and Zhang Tao serves as product director.

Does Butterfly Effect build its own AI models?

No. Butterfly Effect builds no foundation models. Manus is an orchestration layer that routes tasks to third-party models — principally Anthropic’s Claude and fine-tuned Alibaba Qwen — choosing the best model for each step of a task.

Did Meta buy Manus?

Meta announced a deal to acquire Butterfly Effect in December 2025, reportedly valuing it above $2 billion. China’s National Development and Reform Commission blocked the acquisition on 27 April 2026, and by June 2026 Meta had begun unwinding the transaction. The founders are reportedly exploring a buyback.

How much has Butterfly Effect raised?

Butterfly Effect has raised roughly $85 million in disclosed venture funding, most of it in an approximately $75 million Series B led by Benchmark in April 2025 at a valuation near $500 million. Earlier backers include Tencent, HongShan (HSG) and ZhenFund.

Apps

• Manus