Manus Review (2026): Features, Pricing & Verdict
Manus was the autonomous agent that became China’s “second DeepSeek moment” — then the centre of a blocked $2 billion Meta takeover. Here’s an honest look at what it does brilliantly, where it stumbles, and the ownership cloud you can’t ignore in 2026.
In March 2025, a demo video from a little-known Chinese startup racked up hundreds of thousands of views overnight. It showed an AI that didn’t just answer questions — it opened a browser, wrote code, filled a spreadsheet and delivered finished work while the user walked away. That product was Manus (Latin for “hand”), built by Butterfly Effect, the team behind the Monica assistant. The sign-up site crashed within hours, invite codes were resold by scalpers, and more than two million people piled onto the waitlist in a week. Commentators called it China’s “second DeepSeek moment.” Eighteen months on, Manus is a polished, genuinely capable autonomous agent — and also the subject of one of the most consequential AI ownership disputes of the decade. This review covers the consumer Manus app at manus.im.
What Manus actually is
The simplest way to understand Manus is by contrast. When you ask ChatGPT to “research the top 10 project management tools and compare their pricing,” it writes an answer from what it already knows. Give Manus the same prompt and it opens a real browser, visits each tool’s pricing page, extracts the current numbers, builds a spreadsheet and hands you a formatted report. It is an agent<\/strong>, not a chatbot: you describe an outcome, and it plans the steps, executes them and returns finished work.
It does this inside a sandboxed cloud virtual machine<\/strong> with a real browser, terminal and file system. Crucially, the work runs asynchronously<\/strong> — you can kick off a job, close your laptop, and come back later to a completed result. That “leave it to Manus” model is the whole pitch, and when it works it feels genuinely different from babysitting a chat window.
How it works under the hood
Manus is built on a multi-agent architecture<\/strong>. A “Planner” decomposes your request into steps, and “Executor” sub-agents carry out the actions, looping through an analyse → plan → execute → observe cycle until the task is done. At launch, developers who probed the system found it wasn’t a single home-grown model at all — it orchestrated frontier third-party models, principally Anthropic’s Claude<\/strong> and fine-tuned versions of Alibaba’s Qwen<\/strong>, wrapped in Manus’s own coordination layer. That approach let a small team ship a state-of-the-art agent without training a giant model from scratch.
On the GAIA benchmark<\/strong> — a test of real-world, multi-step problem solving — Manus reported scores of around 86.5% on the easiest tier and roughly 70% on the middle tier, which it claimed beat OpenAI’s Deep Research at the time. In 2026 the product exposes its own model tiers in Agent Mode (Manus 1.6 Lite, 1.6 and 1.6 Max<\/strong>), with Max reserved for the longest, most autonomous jobs. The honest read: Manus is fundamentally an orchestration layer over powerful underlying models, which is both the source of its capability and a big reason tasks can get expensive.
Features
Autonomous research & data work
Research is Manus’s strongest suit. Hand it a question and it browses dozens of sources, digs past the top few links, synthesises what it finds and returns a structured, sourced report. Its Code Agent<\/strong> writes and runs Python or JavaScript in the sandbox for data analysis, scraping and automation — so “analyse this CSV and chart the trends” produces an actual file, not just a description.
Web App Builder & Slides
Beyond research, Manus can build and deploy simple web apps and websites<\/strong> from a prompt, and generate AI slide decks<\/strong> for presentations. It’s less of a dedicated app builder than a generalist that happens to ship working prototypes, which is impressive in range even if the polish varies.
Wide Research, My Computer & integrations
The standout power feature is Wide Research<\/strong>, which spins up parallel sub-agents to chew through large, parallelisable jobs far faster than a single agent could. A My Computer<\/strong> desktop app adds local access, and Manus integrates with Slack, WhatsApp and Telegram<\/strong> (plus an Ads Manager integration added in early 2026), so you can trigger and receive tasks where you already work. A lightweight Chat Mode<\/strong> handles quick questions without spending heavy agent credits.
Pricing in 2026
Manus charges for the work the agent does, not per seat or per message: a baseline subscription comes with a pool of credits<\/strong> that every task consumes. The lineup was rebranded in early 2026; figures below reflect the official pricing page as of mid-June 2026, but always confirm before you pay.
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | 300 daily refresh credits, 1 concurrent task, 2 scheduled tasks, Chat Mode and Manus 1.6 Lite in Agent Mode — usable, not a crippled trial |
| Pro (entry) | $20 / mo | 4,000 monthly credits, full Agent Mode (1.6 Lite / 1.6 / 1.6 Max), advanced research, website deployment, slide generation, Wide Research, 20 concurrent and 20 scheduled tasks, beta access |
| Pro (mid) | $40 / mo | 8,000 monthly credits with a 7-day free trial; same feature set as the $20 tier |
| Pro (power) | $200 / mo | 40,000 monthly credits — the realistic floor for heavy, daily agent workflows and the best per-credit value |
| Team | from $20 / seat / mo | 2-member minimum, pooled credits, SSO, usage analytics and admin controls |
| Enterprise | Custom | Centralised billing, compliance and security features — by contacting Manus directly |
Two things to keep in mind. First, monthly credits don’t roll over<\/strong> — daily refresh credits expire after 24 hours, and only purchased add-on credits carry forward (and only while your subscription stays active). Annual billing saves around 17%. Second, the average task runs roughly $2 in credits<\/strong> — about a tenth of what early rivals like OpenAI’s Deep Research cost — but consumption is genuinely hard to predict, and a few big jobs can drain a monthly pool faster than you expect.
The Meta deal and the ownership question
This is the part most feature reviews skip, and it matters more than any single capability. In June–July 2025<\/strong>, Manus quietly moved its headquarters to Singapore and switched its operating entity to Butterfly Effect Pte. Ltd.<\/strong> On 29–30 December 2025<\/strong>, Meta announced a roughly $2 billion acquisition<\/strong>, stating that no continuing Chinese ownership interests would remain after closing.
Beijing pushed back hard. China’s National Development and Reform Commission opened a formal probe in January 2026<\/strong>; the Ministry of Commerce launched a national-security review; and in late March, founders Xiao Hong and Ji Yichao were reportedly barred from leaving China<\/strong>. On 27 April 2026<\/strong>, regulators formally blocked the deal<\/strong>, citing technology-leakage and export-control concerns — even though around 100 Manus staff had already moved into Meta’s Singapore offices. The product still runs normally as a standalone subscription, but the unwind is, by all accounts, still in progress.
For anyone weighing a subscription, this creates three real risks. Roadmap risk<\/strong>: a Meta-owned Manus would have integrated with Meta’s ad and infrastructure stack — that’s now on indefinite hold. Execution risk<\/strong>: founders are travel-restricted and the team is split across Singapore and Meta’s payroll. And data-sovereignty risk<\/strong>: neither the US nor China fully accepts the Singapore entity as legal cover, Meta hasn’t published a unified post-acquisition data-governance statement, and users can’t confirm whether their data has been migrated off legacy servers. Some US state agencies have already restricted the tool, and reports suggest enterprise customers are drifting toward Microsoft and OpenAI. The buying question in 2026 isn’t only feature-versus-feature — it’s whether the company running your agent will exist in a form you trust twelve months from now.
Pros and cons
✅ Strengths
- Genuine end-to-end autonomy — plans and executes multi-step tasks
- Delivers complete artifacts: reports, spreadsheets, slides, web apps
- Deep, thorough autonomous research that digs past the obvious sources
- Runs asynchronously in the cloud — work continues after you log off
- Wide Research parallelises big jobs across many sub-agents
- Usable free tier and low per-task cost (~$2) versus early rivals
- Desktop app plus Slack, WhatsApp and Telegram integrations
❌ Weaknesses
- Credit consumption is fast and hard to predict — budget anxiety
- Uneven reliability on complex or novel tasks; can loop or stall
- Doesn’t ask clarifying questions, so vague prompts misfire
- AI-generated code often needs debugging before production use
- Ownership unresolved after the blocked Meta acquisition
- Data-sovereignty and governance questions left unanswered
- Integrations and ecosystem thinner than ChatGPT’s
Where it falls short
The product weaknesses cluster around predictability<\/strong>. Credits drain quickly and unevenly, so the same kind of task can cost wildly different amounts. Long jobs can be slow, and reliability is patchy: Manus is strong on standard, well-trodden patterns but wobbles on novel business logic, occasionally looping or producing output that needs real cleanup. It also tends to assume rather than ask<\/strong> — feed it a vague prompt and it will confidently build the wrong thing. The practical workarounds are to scope tasks tightly, start with one piece before scaling up, and expect to iterate, especially on code headed for production.
The bigger shortfall is structural. The blocked Meta deal has left Manus in an unusual limbo — effectively US-bought but with China still asserting jurisdiction — and that uncertainty bleeds into data governance and long-term direction in a way no feature can offset. For confidential, client or business-critical work, you need a backup plan. And for anything that can be broken into discrete prompts, tools like ChatGPT or Claude are usually faster, steadier and better supported. Manus earns its place specifically when you need genuinely hands-off, multi-step execution.
Scorecard
How Manus scores across what matters, averaging to 7.4 / 10<\/strong>:
Verdict
Manus genuinely advanced what a consumer AI agent could do. It was among the first to make “give it a goal, walk away, come back to finished work” feel real, and on its best days it still delivers complete, useful artifacts that single-prompt chatbots simply can’t. For open-ended research, data wrangling and prototype-building — especially jobs you can hand off and leave running — it’s one of the most capable agents available, and the free tier makes it painless to try.
It lands at 7.4, not higher, for two reasons. Day to day it’s less predictable than its demos: credits vanish quickly, complex jobs wobble, and it assumes when it should ask. And structurally, the blocked Meta deal has left ownership, roadmap and data governance unresolved in a way no feature list can paper over. Use Manus for what it’s uniquely good at, keep sensitive and business-critical work on a platform whose future you can see, and check the live pricing page before committing. For the China-origin capability story with comparable caveats, see our DeepSeek review, Qwen review and Kimi review; for all-round agents on clearer corporate footing, our ChatGPT review and Claude review are the place to start.
Frequently asked questions
Is Manus free?
Yes — there’s a free tier at $0 with 300 daily refresh credits, 1 concurrent task, 2 scheduled tasks, Chat Mode and Manus 1.6 Lite in Agent Mode. It’s genuinely usable for testing. Full agent tools, more powerful models and bigger credit pools require a paid plan from $20/month.
What does Manus actually do?
It’s an autonomous AI agent. You give it a goal and it independently plans the steps, browses the web, writes and runs code, analyses data, and delivers finished outputs — sourced reports, spreadsheets, slide decks, even simple web apps — inside a cloud sandbox that keeps working after you disconnect.
Is Manus owned by Meta?
No. Meta announced a roughly $2 billion acquisition in December 2025, but China’s regulators formally blocked it on 27 April 2026, citing technology-leakage and export-control concerns. Manus continues as a standalone service run from Singapore, with long-term ownership unresolved.
Is Manus safe and private to use?
Treat it with caution for sensitive data. After the blocked deal there’s no unified data-governance statement, the Singapore entity’s legal protection is contested by both the US and China, and some US agencies have restricted the tool. It’s fine for general, non-confidential work; keep client, personal or business-critical data on a platform you trust.
How much does Manus cost per task?
On average roughly $2 in credits per task — about a tenth of early competitors like OpenAI’s Deep Research. But credit consumption is unpredictable and scales with task complexity, so heavy or long-running jobs can cost considerably more.
Is Manus better than ChatGPT?
For hands-off, multi-step autonomous execution — research 10 competitors, extract their pricing and compile a spreadsheet — Manus is purpose-built and often goes further. For tasks you can break into individual prompts, ChatGPT or Claude is usually faster, more reliable and backed by a deeper ecosystem.
Reviewed June 2026 by AINewsAndUpdates.com. Models, features, pricing and corporate ownership are changing fast — always confirm current details on the official site before relying on it.