- Manus AI launched via invite-only preview on March 6, 2025, built by Chinese startup Butterfly Effect (Monica.im) and pitched as one of the first 'fully autonomous' general-purpose agents.
- It reported topping OpenAI's Deep Research across all three GAIA benchmark difficulty levels — but reviewers found frequent task failures, crashes, and looping once the hype settled.
- Butterfly Effect raised roughly $75 million led by Benchmark at a valuation near $500 million, then moved its headquarters toward Singapore amid US-China scrutiny.
- Paid tiers now run roughly $39 to $199 a month, replacing the scarce invite codes that were reselling on Chinese secondary markets for a premium.
What's on the Table
$500 million. That's the valuation attached to a startup most of the tech industry hadn't heard of before March 2025. According to Google News, Manus AI arrived that month as an invite-only preview from Butterfly Effect, the company behind the Monica.im browser assistant, and it didn't pitch itself as another chatbot — it claimed to be a fully autonomous agent that could take a multi-step task, plan it, and execute it end-to-end with minimal human babysitting.
The proof point Butterfly Effect leaned on was the GAIA benchmark, a General AI Assistants test co-created by Meta AI and Hugging Face. Manus reported higher pass rates than OpenAI's Deep Research across all three GAIA difficulty levels, and that claim — paired with scarce invite codes reselling for inflated prices on Chinese secondary markets — produced the kind of viral scramble the industry hadn't seen since DeepSeek's debut. Commentators explicitly drew that comparison, framing Manus as evidence that Chinese labs were shipping agentic products, not just larger chatbots.
Side-by-Side: How They Differ
Here's the part that got lost in the launch-week noise: Manus is not a new foundation model. Reporting indicates it's an orchestration layer that routes tasks across existing models — Anthropic's Claude alongside fine-tuned open models like Alibaba's Qwen — inside a multi-agent framework that executes inside a cloud sandbox. Skeptical analysts summed it up bluntly as 'a wrapper done exceptionally well,' which is a fair description of the pattern: a planner agent breaks a request into subtasks, hands them to specialized sub-agents or tools (browser, code executor, file system), and stitches results back together.
That's meaningfully different from how OpenAI's Deep Research and Operator work, or how Anthropic's 'computer use' and Google's Project Mariner approach the same problem — each ties agentic behavior more tightly to its own first-party model rather than orchestrating across multiple vendors' weights. The funding math tells its own story too: Butterfly Effect closed roughly a $75 million round led by Benchmark at a valuation near $500 million, then relocated its headquarters toward Singapore as US-China tensions around AI intensified — a dynamic explored further in White House's Chinese AI IP Theft Claim: What We Know. Pricing moved from invite scarcity to a straightforward Starter/Pro structure once broader access opened.
Chart: Manus AI's reported Starter and Pro monthly pricing tiers after the invite-only phase ended. The roughly fivefold jump between tiers signals Butterfly Effect is pricing Pro access for sustained, agent-heavy workloads rather than casual use — our read is that's a pricing structure built for compute cost recovery on long autonomous runs, not a typical SaaS seat-based ladder.
Photo by Priyank Pathak on Unsplash
The AI Angle
Manus is a symptom of a bigger 2025-into-2026 pivot: the industry moving from reactive chatbots to agentic AI — systems that plan and execute multi-step tasks using tools, browsers, and code with minimal prompting per step. OpenAI's Deep Research and Operator, Anthropic's computer-use capability, and Google's Project Mariner are all racing the same territory Manus staked out first. For financial management specifically, this is the layer enterprises are watching for autonomous research, reconciliation, and reporting workflows — the pattern matters more than any single vendor's benchmark win.
Which Fits Your Situation
Reviewers reported crashes and looping in hands-on testing — a classic tool-call loop failure mode where the planner agent re-invokes the same sub-agent without converging on a result. That's expensive in both latency and token cost, and it's exactly where multi-agent orchestration breaks in production.
The GAIA numbers favor Manus on paper, but benchmark wins don't always survive contact with messy, real-world context windows — run your own task set before trusting a vendor's leaderboard claim.
The relocation amid US-China scrutiny is a governance signal worth flagging to compliance teams before piloting Manus in regulated financial workflows.
Frequently Asked Questions
What is Manus AI and how does it work?
Manus AI is a general-purpose autonomous agent from Chinese startup Butterfly Effect (Monica.im), publicly unveiled via invite-only preview on March 6, 2025. It orchestrates multiple existing models — including Anthropic's Claude and fine-tuned open models like Alibaba's Qwen — in a multi-agent framework running inside a cloud sandbox to complete multi-step tasks with minimal human oversight.
Is Manus AI better than ChatGPT or OpenAI Deep Research?
At launch, Manus reported outperforming OpenAI's Deep Research across all three GAIA benchmark difficulty levels. But hands-on reviewers reported frequent task failures, crashes, and looping, leading some outlets to frame the launch as overhyped rather than a clean technical win.
Who created Manus AI and where is the company based?
Manus AI was created by Butterfly Effect, the Chinese startup behind the Monica.im browser assistant. The company raised roughly $75 million led by Benchmark at a valuation near $500 million and later relocated its headquarters toward Singapore amid US-China AI scrutiny.
How much does Manus AI cost?
After the invite-only phase, Butterfly Effect introduced paid subscription tiers reportedly ranging from about $39 to $199 a month across Starter and Pro plans, replacing the scarce invite codes that had been reselling at a premium on Chinese secondary markets.
Disclaimer: This article is editorial commentary based on publicly reported information and does not constitute financial, investment, or product-testing advice. Research based on publicly available sources current as of July 24, 2026.