Manus review: an AI agent for finished work, paid in credits

Manus is a general-purpose AI agent that takes a task and delivers a finished result: a research report, a website, a slide deck, a short video or a simple game. It works in its own cloud environment, so you describe what you want and come back to the output. It’s a good fit for people who need deliverables with a clear finish line. It’s a weaker fit for open-ended errands, and its credit-based pricing makes it hard to know in advance what a task will cost.

Who Manus is for

Manus suits people whose work ends in a file or a link:

• Marketers and founders who need landing pages, decks and short product videos.
• Consultants and analysts who need a researched report with sources gathered from many pages.
• Small teams that want recurring automations, such as a weekly summary built from Gmail, Slack or Notion.

Users on Reddit tend to agree on where it works: concrete multi-step tasks with a clear end, rather than vague, long-running goals (Ciela’s summary of Reddit threads). If what you want is an assistant that handles your inbox and bookings every day, Muse or Grok Bot fit better.

What’s new

Manus went viral as an invite-only beta in March 2025, when invite codes were resold for over $1,000 (Wikipedia). Since then:

• The Meta deal fell apart. Meta agreed to buy Manus in December 2025. China investigated, and in April 2026 ordered the deal unwound. Manus now operates independently from Singapore, and says it was required to delete some user data as part of the split (TechCrunch).
• Manus 2.0 (September 28). A new agent harness called Cascade, which Manus says used 23.2% fewer tokens and cost 32% less in one tested setup; Automations that start from events in connected apps, not just on a schedule; Cloud Computers you can buy that keep running when your laptop is closed; and Manus Studio on desktop, with a Video Editor and Game Dev workspace (Manus).
• Cue. A separate app, launched alongside 2.0, where each personal agent gets its own email address, phone number, wallet and computer. It still requires an invitation code (Manus help center) and is a different product from Manus itself (The Next Web).
• Manus in your chats. Manus Agents now work in Telegram for every user on every plan, running the full agent rather than a lighter chatbot, and an automation can deliver scheduled summaries straight into the chat (Manus). You can also message it in Slack channels and DMs and get files back there (help center). That moves Manus closer to an assistant that reports to you, not just a tool you visit.
• New funding (October 8). More than $500 million. Manus didn’t disclose a valuation; in September it was reported to be in talks to raise at $4 billion (TechCrunch). The agent Meta tried to buy now competes with Meta’s own Muse.

Real use cases

The task box on the homepage gives the clearest picture: slides, websites, games and video, alongside open research. In practice:

• Research reports. “Compare the five biggest project management tools for a 20-person team” becomes a sourced document.
• Websites and web apps. A landing page or a small internal tool from a description.
• Slide decks from a brief or a document.
• Short videos and games, which Manus Studio lets you edit instead of regenerating from scratch.
• Automations. A report that rebuilds itself when a new email arrives or a Notion page changes.

How Manus compares

vs Grok Bot. Grok Bot runs long-lived, named Bots inside your accounts: routines and inbox work. Manus is built around projects with an output. Grok Bot has no free plan and starts at $30 a month; Manus has a free plan and paid plans from $20. Choose Grok Bot for ongoing chores and Manus for deliverables.

vs Muse and Instinct. Those are consumer errand agents: shopping, bookings, email. Manus doesn’t really compete with them. Cue, Manus’s separate app, is its answer to them.

vs OpenClaw. OpenClaw is free, open source and runs on your own computer with the model you choose, but expects command-line skills and careful security settings. Manus is hosted and needs no setup. If you want control and are technical, OpenClaw. If you want a finished deck by this afternoon, Manus.

vs dots. OpenAI’s always-on agent is the closest match in what it produces: it uses Codex and ChatGPT Work to research, analyze data, prepare documents and build software on its own cloud computer, and it works from your ChatGPT memory. It’s included only from the $100 ChatGPT Pro plan and you get one dot. Manus starts free, bills per task, and its editors for video and games go further. If you already pay for ChatGPT Pro, try your dot first.

vs Lindy. Lindy is an assistant for the daily work around deliverables: triaging email, scheduling, recording and prepping meetings, with daily briefs over iMessage or Slack. Manus makes the deliverable itself. Small teams sometimes need both; if you have to pick, decide whether your bottleneck is your inbox or your output.

Pricing advice

Manus charges in credits, and the number a task uses depends on how complex it turns out to be (Manus help center). You can’t know a task’s cost before it runs, and runs that go wrong can still use credits; Manus refunds only tasks that fail because of a technical issue on its side. So:

• Start on the free plan. Daily free credits only work in the Lite mode, so the free plan shows you the workflow, not the full agent.
• Give precise briefs. Vague prompts are what lead to long, expensive runs. Say exactly what the output should be.
• Pick the $20 Pro plan first. All three paid individual plans are called Pro and differ only by credit amount. Move up only when you run out every month. The $40 plan has a 7-day free trial if you want to test the full agent first (pricing).
• Annual billing saves 17%, but only commit after a month or two of real use.
• Budget for extras. Cloud Computers are paid separately. Monthly plan credits don’t roll over; add-on credits carry over only while you keep a paid subscription.

Privacy

Manus is weaker here than most rivals:

• A training opt-out is listed only for the Team plan. Manus doesn’t state an in-app opt-out for individual plans, and we couldn’t confirm whether it trains on individual users’ data by default.
• A change of hands left a mark. Manus said it was required to delete some user data as part of its split from Meta (TechCrunch), a reminder that ownership and jurisdiction can change what happens to your data.
• Regulatory uncertainty. The company was built in China, moved to Singapore, and has been caught between US and Chinese regulators (Wikipedia).

For company work with sensitive data, use the Team plan with its opt-out, or keep confidential material out of Manus.

Getting good results

Most complaints about Manus come down to wasted credits on runs that didn’t produce what the user wanted. A few habits help:

• Describe the finished output. “A 10-slide deck for investors, with a market size slide and three competitor slides” works better than “make a pitch deck”.
• Give it the sources you trust, such as links or files, instead of letting it search widely.
• Break big jobs into steps. A research report first, then a deck built from it, rather than one long run that can go wrong halfway.
• Edit instead of regenerating. For video and games, Manus Studio’s editors let you fix parts of the result without paying for a full new run.
• Test automations by hand first. Run the task once yourself before setting it to trigger on every new email.

When Manus is the wrong tool

Skip Manus if you mainly want a daily assistant for email, bookings and shopping; Muse is free and built for that. Skip it if you need predictable monthly costs, since credits make that hard. And skip it for confidential client material unless you’re on the Team plan with its training opt-out.

Verdict

Manus is one of the best agents for turning a brief into a finished deliverable. Manus 2.0 made it faster and more useful, especially with event-triggered automations and editable video and games. If your work produces reports, sites or decks, it’s worth trying, and the free plan makes that easy.

The weak points are cost and privacy. Credits make spending hard to predict, failed runs still cost you, and individual users get no clear training opt-out. Write precise briefs, start on the cheapest Pro plan, and keep sensitive material on the Team plan or out of it. For other options, see our Manus alternatives or our guide to the best AI agents.