Manus AI Explained
What the general autonomous agent does, what it costs, and where it breaks down
Manus AI is a general-purpose autonomous agent that can take a task from plan to completion inside a cloud-hosted virtual computer. It browses live websites, writes and runs code, edits files, and creates deliverables such as slide decks, spreadsheets, and small web apps, all without a human clicking through each step.
Unlike most chatbots, which simply answer questions, Manus does the work behind them. Give it a goal, say, "Research these five competitors and build a comparison deck", and it will open a browser, visit relevant sites, extract the data, and return a finished file.
This guide explains how Manus AI works, what it costs under the current credit system, how it compares with Hermes Agent, Devin, and OpenAI Operator, and where it tends to fall short on real-world tasks.
Manus AI Scheduled Automations and Plan Limits
Manus AI runs tasks as an autonomous agent inside a cloud virtual computer. Scheduled Automations run tasks at set times as part of the broader Automations feature. You set up a scheduled task by opening Automations in the sidebar, selecting Create, and choosing Schedule.
Step-by-step instructions for Wide Research and Automations are in our guide on how to use Manus AI.
Your plan tier sets how many tasks run at once and how many you can schedule. The Free plan includes 300 daily refresh credits, one concurrent task, two scheduled tasks, and access to Manus 1.6 Lite only. Paid tiers expand capacity to 20 concurrent tasks and 20 scheduled tasks. If those limits are too tight for your workload, see our breakdown of Manus AI alternatives.
What Is Manus AI?
Manus AI is an autonomous agent platform built by the startup Monica, designed to complete open-ended, multi-step tasks with minimal human input.
Instead of a single model answering one prompt, Manus AI runs a central executor agent that breaks a goal into a plan, then calls specialized sub-agents and tools to carry it out: a browser agent for web work, a code agent for scripts, and a file agent for documents.
Every task runs inside a sandboxed cloud computer, not on your local machine. That virtual computer has its own file system, terminal, and headless Chromium browser, so Manus AI can install packages, run shell commands, and click through real web pages the same way a person would.
- Plans multi-step tasks from a single high-level prompt
- Runs in an isolated cloud virtual computer, not your device
- Combines browsing, code execution, and file generation in one session
- Ships finished outputs (decks, sheets, sites, reports), not just text answers
Who Owns Manus AI? The Meta Deal and Its Reversal
Manus is an independent company again, led by its founding team. Meta announced an agreement to acquire Manus in December 2025 for a valuation reported by TechStartups and The Next Web at about $2 billion. China's National Development and Reform Commission (NDRC) ordered the transaction reversed on April 27, 2026, citing foreign-investment and technology-export rules.
Update, September 2026: Manus is in talks to raise about $500 million at a $4 billion valuation, TechCrunch reported on September 18. The same TechCrunch report says Manus is considering a restructuring to prepare for an initial public offering (IPO) in Hong Kong. Early investors and backers reportedly helped Manus buy back its shares at about a $2 billion valuation.
During the system split in mid-2026, some accounts experienced temporary access interruptions, and users in certain regions were asked to back up data created on or after December 29, 2025, and some of it was deleted. Manus established an account restoration portal with no closing deadline for affected users.
Manus operates from Singapore and now calls itself an independent agent lab. Chief executive officer (CEO) Xiao Hong leads the business alongside co-founders Zhang Tao and Ji Yichao. Subscribers who were not affected do not need to do anything.
- December 29, 2025: Manus announced it was joining Meta. The price was later reported at about $2 billion.
- April 27, 2026: China's National Development and Reform Commission (NDRC) prohibited the investment and ordered the deal unwound.
- June 2026: Meta separated its internal systems from Manus, while the founders reportedly sought outside capital to repurchase the company.
- September 1, 2026: Manus issued an official announcement confirming the formal resumption of independent operations.
- September 18, 2026: TechCrunch named IDG Capital, Boyu Capital and battery maker Contemporary Amperex Technology (CATL) as potential investors in the $500 million round Manus is discussing.
- September 2026: Manus is not publicly traded, so there is no Manus stock to buy. Manus is reportedly considering a restructuring to prepare for a Hong Kong IPO.
How Manus AI Works
Manus AI works by splitting one goal into a visible task plan, then executing each step with the right tool before assembling a final deliverable.
When you submit a prompt, an orchestrator agent writes out a task list you can watch update in real time. It then hands pieces of that list to sub-agents: one opens a browser tab and navigates a site, another writes and tests code in the terminal, another edits a spreadsheet or slide file.
The browser agent is worth calling out separately. It renders pages visually, so it can read charts, fill multi-field forms, and click buttons that a text-only scraper would miss. That's the main reason Manus AI handles research and data-extraction tasks that trip up simpler agents.
Manus also has a desktop app for Mac and Windows that works with folders you authorize on your own computer.
- Step 1: Orchestrator agent turns your prompt into a visible task plan
- Step 2: Sub-agents execute steps using browser, terminal, and file tools
- Step 3: The agent self-checks output before marking a step done
- Step 4: You get a finished file, live link, or report, plus a replay of the steps taken
What Models and Modes Does Manus AI Use?
Manus AI runs on its own Manus 1.6 model family, offered in Lite, Standard, and Max tiers depending on your plan.
Manus launched Manus 2.0 on September 28, 2026, and the Manus help center calls Manus 2.0 the current version. Other Manus help-center articles still name the modes Manus 1.6 Lite, Manus 1.6 and Manus 1.6 Max, so the mode names in your account may differ from the names on this page.
Free accounts get Chat Mode plus Manus 1.6 Lite in Agent Mode. Paid Pro and Team accounts unlock Manus 1.6, Manus 1.6 Max, and Chat Mode together, along with a Wide Research feature that fans a single research task out across many sources at once.
Manus AI does not publish which base foundation models sit under the 1.6 family, and that mix can change between releases. Treat any specific claim about the underlying model as unverified unless the vendor states it directly.
For step-by-step use of Wide Research, Automations and Playbooks, see how to use Manus AI.
- Chat Mode: conversational, no autonomous task execution
- Agent Mode: full autonomous execution with browser, code, and file tools
- Manus 1.6 Lite: fastest and cheapest per credit, free-tier default
- Manus 1.6 / 1.6 Max: higher-reasoning tiers on paid plans
- Wide Research: parallel multi-source research on Pro and Team plans
How Much Does Manus AI Cost?
Manus AI runs on a credit system layered under three plans: Free, Pro, and Team, with credits consumed per task based on complexity rather than a flat per-task fee.
The Free plan gives 300 daily refresh credits, one concurrent task, two scheduled tasks, and Manus 1.6 Lite only. Pro plans start around $20 per month for roughly 4,000 monthly credits, with a higher tier near $40 per month for about 8,000 credits. A $200 per month tier adds 40,000 credits. The paid tiers unlock 1.6, 1.6 Max, and 20 concurrent and scheduled tasks. Team plans start near $20 per seat per month with a pooled credit balance shared across the team, plus SSO and usage analytics.
Annual billing on paid plans currently runs about 17% cheaper than monthly. These figures move often on agent platforms - verify current numbers on Manus's pricing page before quoting a client.
- Free: $0/mo, 300 daily credits, 1 concurrent task, Lite model only
- Pro (lower tier): from $20/mo, ~4,000 monthly credits, full model access
- Pro (higher tier): from $40/mo, ~8,000 monthly credits, 7-day trial
- $200/mo tier: 40,000 monthly credits ($167/mo billed yearly)
- Team: from $20/seat/mo, pooled credits, SSO and access control
How Does Manus AI Differ from Hermes Agent, Devin, and OpenAI Operator?
Manus AI is a general-purpose agent for research, documents, and light app-building, while Hermes Agent, Devin, and OpenAI Operator each specialize in a narrower job.
Devin, built by Cognition, is a coding-focused agent. It works inside a dev environment with its own shell, editor, and browser, and it's judged mainly on how well it ships pull requests and fixes bugs in an existing codebase. Manus AI can write code too, but it's built for the whole task, from web research to the final deliverable, not for living inside a repo.
OpenAI Operator (and its successor agent tooling) is closer to Manus AI in that it also drives a browser to complete web tasks like booking or form-filling. The difference is scope: Operator stays tightly browser-centric, while Manus AI pairs browsing with a full code and file sandbox in the same session.
Hermes Agent, from Nous Research, targets a different buyer entirely: teams that need an agent they can self-host or audit, not a hosted black box. If data residency or model transparency matters more than convenience, Hermes Agent is the one to evaluate against Manus AI, not swap in for it.
For how Manus compares with Anthropic's assistant, see whether Manus AI is better than Claude.
For OpenAI's always-on agent and the other options in this category, see the best alternatives to ChatGPT dots.
- Manus AI: general agent, browser + code + files, hosted cloud sandbox
- Devin: coding-specific agent, lives inside a dev repo and PR workflow
- OpenAI Operator: browser-first task automation, narrower scope than Manus AI
- Hermes Agent: self-hostable/auditable agent stack, different buyer than Manus AI
What Are the Limits and Failure Modes of Manus AI?
Manus AI's main limits are cost predictability, task reliability on long-running jobs, and dependence on the sites it browses.
Cost per task is not fixed. Credits burn faster on tasks with many browser steps or large file generation, so two prompts that look similar in length can cost very different amounts. Budget in credit ranges, not per-task estimates.
Long multi-hour tasks are where Manus AI is most likely to drift: it can misread a step earlier in the plan and keep building on that error instead of catching it, producing a deliverable that looks finished but is wrong in a way that takes a careful read to spot. It also inherits the fragility of any browser agent, if a site changes its layout, adds a CAPTCHA, or sits behind a login, the task can stall or silently skip a step.
The practical failure mode we'd flag for a business buyer isn't a crash, it's a confidently wrong output. A research summary with one fabricated or misattributed data point is worse than an obvious error, because it passes a quick skim. Any Manus AI output headed to a client or a decision-maker needs a human check before it ships.
- Credit cost per task varies with step count and file size, not fixed
- Long tasks can drift on an early misread step without flagging it
- Browser tasks stall on CAPTCHAs, logins, or changed page layouts
- Confidently wrong output is the real risk, not visible crashes
- Concurrent/scheduled task caps (1 free, 20 on Pro) limit batch use
Who Should Use Manus AI?
Manus AI fits teams that need finished research, reports, or light web apps produced without a human driving every step, and can tolerate reviewing the output before it goes out the door.
It's a strong fit for market research, competitive teardown decks, spreadsheet-heavy data pulls, and quick internal tools where speed matters more than production-grade code. It's a weaker fit for anything that needs an audit trail on where the underlying model runs, given the model-transparency gap versus a self-hosted stack like Hermes Agent, or for tasks that need guaranteed, repeatable output on every run.
When we scope agent rollouts for clients through our AI workflow audits, the pattern that holds across tools like Manus AI is consistent: the agent earns trust fast on drafting and research work, and needs a human checkpoint before it touches anything client-facing or financial. That checkpoint isn't optional friction, it's what keeps a fast tool from shipping a fast mistake.
If Manus is not the fit, compare the best Manus AI alternatives, including free ones.
Frequently Asked Questions
- Yes, Manus AI has a free plan with 300 daily refresh credits, one concurrent task, and access to Chat Mode plus the Manus 1.6 Lite model in Agent Mode. Paid Pro and Team plans add higher credit allowances, faster models, and more concurrent tasks. If you want a self-hosted option instead, see how to install and run OpenManus for free.
- Yes, Manus AI is a Chinese-founded, Singapore-based company. Manus AI resumed independent operations on September 1, 2026, after China's National Development and Reform Commission (NDRC) ordered Meta's $2 billion acquisition withdrawn on April 27, 2026. Chief executive officer (CEO) Xiao Hong leads Manus AI alongside co-founders Zhang Tao and Ji Yichao.
- No, Manus AI is not publicly traded, so there is no stock to buy. As of September 2026, Manus is reportedly in talks to raise $500 million at a $4 billion valuation. Manus is also reportedly considering a restructuring to prepare for an initial public offering (IPO) in Hong Kong.
- Manus AI plans and executes multi-step tasks on its own: browsing websites, running code, editing files, and building finished deliverables like slide decks, spreadsheets, and simple web apps, all inside a cloud sandbox rather than your local machine.
- Manus AI runs tasks in an isolated cloud sandbox, and Team plans add SSO and a data-training opt-out, but the company does not publish full model-transparency details the way a self-hosted option does. Review sensitive-data policy before routing client or financial work through it, and always human-check output before it ships externally.
- ChatGPT and Claude are chat-first assistants that can call limited tools inside a conversation. Manus AI is agent-first: it takes one high-level goal, runs a multi-step plan on its own across browsing, code, and files, and returns a finished deliverable instead of a conversational answer. For the full comparisons, see whether Manus AI is better than ChatGPT and Manus vs Claude.
- There is no fixed per-task price. Manus AI runs on a credit system where cost depends on task complexity, step count, and file generation, so a simple lookup costs far fewer credits than a long research-and-build job. Check Manus's pricing page for current credit-to-dollar ratios before budgeting a project.
- No. Manus AI can write and run code, but it's built for general tasks spanning research, documents, and light apps. A dedicated coding agent like Devin or OpenAI Codex is the better fit for working inside an existing codebase and shipping pull requests.
- The most common failure is a deliverable that looks complete but contains a wrong or fabricated detail from an early misread step, not a visible crash. Long, browser-heavy tasks are most at risk. Always review Manus AI output before using it in a client-facing or financial decision.
Not sure if Manus AI fits your workflow?
Autonomous agents like Manus AI can cut real hours off research and reporting work, but only when the task, the review step, and the cost ceiling are scoped up front. We'll map where an agent like this actually fits your team's workflow, and where it needs a human in the loop.
Book a Consultation