The Truth About Manus AI: The $2B Deal China Blocked, and What It Actually Does
The viral demo was real. So was the regulatory order that blew up its acquisition.
Manus launched as an invite-only beta on 6 March 2025, and the demo genuinely was as dramatic as it looked: an AI agent opening a browser on its own, comparing property listings, building a spreadsheet, checking flights, and returning a finished travel itinerary — no human clicking anything in between. The waitlist crossed six figures within 48 hours, and invite codes reportedly changed hands for up to 50,000 yuan.
What most coverage outside China hasn't caught up to: Meta agreed to acquire the company for more than $2 billion — and on 27 April 2026, China's National Development and Reform Commission blocked the deal outright, ordering both parties to withdraw, with no detailed public explanation given.
The product is still running. The question anyone evaluating it now has to ask is genuinely new: what does this company look like in a year?
📋 In This Guide
What It Actually Does Well
Built by Butterfly Effect, a startup founded in China and headquartered in Singapore since mid-2025 — the same team behind Monica, an earlier browser-based AI assistant. Unlike a chatbot, Manus doesn't wait for your next message: give it a goal in plain language, and it plans the steps, opens a browser, writes and runs code, and returns a finished result inside a cloud-hosted virtual computer.
Agent Mode is the feature the whole product is built around — genuine autonomy on multi-step work. Chat Mode handles smaller requests that don't need the full sandbox. Advanced and Wide Research modes, consistently praised across reviews, pull from multiple sources and return structured reports, comparison tables and executive summaries rather than a single chat reply.
A newer Web App Builder generates full websites and apps with a built-in database, Stripe integration and basic SEO scaffolding — genuinely more than a static mockup, though every review consistently frames the output as a fast prototyper, not a finished-product builder.
The consistent summary across independent reviews: it's genuinely impressive for individual research and web-based tasks — the part of the original demo that actually holds up.
The Credit System — The Real Story
This is where "the truth" earns its place in the title. Pricing runs on credits: free tier with roughly 1,000 starter credits plus a 300 daily refresh, Standard at $20/month for 4,000 credits, a $40 tier for 8,000, and $200 for 40,000.
The genuinely damning part isn't the price itself — it's that there's no cost estimate shown before a task runs, only after. One detailed hands-on test ran the identical prompt — a comparison of five tools' pricing plans — through three different tiers of the model and logged the credit meter after each run. The results: 42, 69, 274 and 94 credits for functionally similar tasks. The reviewer's own conclusion: "the price signal did not track the quality signal." Paying more didn't reliably buy better output, and there was no way to know the cost in advance either way.
That unpredictability shows up in public sentiment too: Manus AI's Trustpilot rating sits around 1 out of 5, with reviews centred on billing complaints specifically rather than the product's core capability. That's a genuinely useful distinction — the anger is about not knowing what something will cost, not about whether it works at all.
⚡ The Product Has a Real Ceiling, Not Just a Pricing Problem
Sixteen months after launch, reviewers converge on the same specific gaps.
Where It Genuinely Falls Short
Beyond the credit issue, three limitations recur across nearly every independent review checked:
It's built for solo use, not teams. No shared team workspace, no persistent memory across sessions, and only a small set of external connectors — a real ceiling for anyone hoping to run it as shared infrastructure rather than a personal tool.
Reliability drops under real conditions. Reviewers report complex tasks failing mid-stream, credit consumption behaving unpredictably, and noticeably worse reliability during peak usage hours.
Generated code is prototype-grade, not production-grade. Naive database schemas and minimal error handling are named specifically — usable for a quick demo, not something to ship without a developer reviewing and hardening it first.
The consistent conclusion across sources, worth taking at face value: Manus is a strong research and orchestration tool, not a replacement for purpose-built coding agents or for a human carefully reviewing what it produces.
The Blocked Deal, and Why It Matters
Worth understanding what this actually changes for anyone evaluating the tool today: a blocked acquisition means the corporate future is genuinely uncertain, not that the product itself is unsafe or about to disappear. It's still running, and the waitlist history and usage numbers suggest real, continued demand.
But for anyone considering building a real workflow or business process on top of it — as opposed to using it for personal research or experimentation — the honest caveat is that ownership, roadmap and long-term investment are all open questions in a way they weren't before the blocked deal. That's a different kind of risk from "will this feature work today," and it's worth weighing separately.
Who This Is Actually For
The honest match, drawn from the consistent pattern across reviews: good for autonomous research, data gathering and quick prototypes, particularly on the entry $20 tier where the credit risk is bounded. A harder sell for predictable costs, everyday production reliability, or shippable app output you plan to depend on without a developer reviewing it first.
The practical framing worth adopting: treat it as a tool for research and experimentation, not production work you need to rely on — a distinction that holds regardless of how the corporate situation eventually resolves.
Frequently Asked Questions
Did Meta really try to acquire Manus AI?
Yes. Meta reportedly agreed to acquire the company for more than $2 billion, but China's National Development and Reform Commission blocked the deal on 27 April 2026, ordering both parties to withdraw.
Is Manus AI still available and being developed?
Yes, the product continues running and being updated, including a newer Web App Builder feature. The blocked acquisition creates corporate uncertainty, not an immediate shutdown risk.
Why does Manus AI have such a low Trustpilot rating?
Around 1 out of 5, driven largely by billing complaints tied to its credit-based pricing, where costs aren't shown before a task runs and can vary significantly for similar tasks.
Can Manus AI build production-ready apps?
Not reliably. Its Web App Builder produces functional prototypes, but reviewers consistently note naive database schemas and minimal error handling — a starting point for a developer to harden, not shippable output on its own.
Is Manus AI good for team collaboration?
No, it's built for solo use — no shared team workspace and no persistent memory across sessions, a real limitation for anyone hoping to deploy it as shared infrastructure.
How much does Manus AI actually cost?
Credit-based: a free tier with roughly 1,000 starter credits plus daily refresh, Standard at $20/month for 4,000 credits, up to $200/month for 40,000 credits — but no cost estimate is shown before a task runs.
The Takeaway
The viral demo was real, and Agent Mode genuinely does what it showed — autonomous research and multi-step web work with no hand-holding. The blocked $2 billion Meta acquisition is the part of the story most coverage outside China has missed, and it's worth knowing before you build anything you'd depend on.
The credit system is the part most reviews underweight relative to how much it actually matters — no upfront cost estimate, and one documented test showing the same task costing wildly different amounts depending on tier, with quality not reliably tracking price. Use it for research and prototypes where a bounded budget covers the risk. Think twice before betting production work on it until the pricing model, and the company's ownership, both settle down.
🚀 Stay Connected With Simple AI Tools
Honest AI tool coverage — the capability and the risk, both.
👇 💬 Drop your comment below and let us know your thoughts! ✨
Comments
Post a Comment