Butterfly Effect

Manus

Provider

Butterfly Effect

Price

$0-200/mo (credit-based)

Platforms

Web, iOS, Android

Launched

Mar 2025

Updated June 2026

Manus is an autonomous, general-purpose AI agent built by Butterfly Effect, the Singapore-headquartered company founded by Xiao Hong. Where a chatbot answers, Manus acts: you describe an outcome, and it plans the work, opens a cloud computer with a browser and a code sandbox, and returns a finished artefact — a research report, a slide deck, a working website, a spreadsheet model — rather than instructions for you to follow. It launched in invite-only beta on 6 March 2025, and its demo video drew more than a million views in twenty hours, briefly making invite codes a resold commodity (Wikipedia).

The defining technical fact about Manus is that it does not train its own frontier model. It orchestrates third-party models — principally Anthropic’s Claude and fine-tuned versions of Alibaba’s Qwen — and routes each step of a task to whichever model fits, a design Manus calls multi-model dynamic invocation (MIT Technology Review). Its own product names — Manus 1.6, 1.6 Max, 1.6 Lite — are agent tiers, not foundation models.

One thing to know before you subscribe: ownership is unsettled. Meta announced a deal to acquire Manus in December 2025, reportedly valuing it above $2 billion, but China’s antitrust regulator blocked the acquisition on 27 April 2026 and the matter remains unresolved as of 18 June 2026 (BBC, Wall Street Journal). The product runs normally; the corporate question is covered in full on the Butterfly Effect provider page.

Quick stats

TRY MANUS →

Pricing breakdown

Manus does not charge per seat or per message. It charges for work done: a flat subscription buys a monthly pool of credits that each task draws down based on its complexity. This is the single most important thing to understand about Manus, and the most common source of complaints — a single deep-research run can quietly consume 500-900 credits, and complex automations have been reported to clear 1,000+ in one go (Fello AI).

The 2026 lineup has five price points across three plans. Manus rebranded its older Basic/Plus/Pro tiers in early 2026; figures below are current as of May 2026 and credit allocations change, so verify on the official pricing page.

Annual billing saves 17% across every paid tier (Fello AI). All paid pricing is effectively linear at roughly $0.005 per credit, so the only reason to move up a tier is needing more credits, not a better rate.

Free — $0

Genuinely usable for testing, not a crippled trial. You get 300 daily refresh credits (reset every 24 hours) plus a 1,000-credit one-time starter pack, 1 concurrent task, 2 scheduled tasks, Chat Mode, and Manus 1.6 Lite in Agent Mode. The catch: 300 credits is roughly one moderate slide deck or a few short queries, and a single deep-research run can drain the daily pool in one shot. Free is for evaluation, not shipping.

Pro — $20 / $40 / $200 per month

The Pro plan ladders up purely on credits: $20 for 4,000, $40 for 8,000, $200 for 40,000. All three unlock the full model set (1.6 Lite, 1.6 and the flagship 1.6 Max), 20 concurrent tasks, 20 scheduled tasks, Wide Research, slide generation and website/app deployment. The $40 tier is the only one with a 7-day free trial, which makes it the sensible place to start: a real week to learn your workflow’s credit burn before you drop to $20, stay at $40, or jump to $200.

Subscribe to Pro →

Team — from $20/seat/month

Built for small teams (2-seat minimum). Each seat ships with 8,000+ monthly credits, 300 daily refresh credits, a shared credit pool any teammate can draw from, SSO, access controls and usage analytics. It is the only plan with central billing and admin visibility into who is spending credits on what. Larger or enterprise deployments are handled via contact sales.

How credits are consumed

Credits are spent only while a task is actively running — storage and idle time are free. Three things drive the cost (Manus help centre): the LLM tokens used for planning and output, the cloud virtual machine that runs the browser, files and code, and any third-party APIs the task calls. As a rough guide from user reports:

• A simple chat answer: 10-50 credits
• A slide deck that browses for images: a few hundred credits
• A deep autonomous research run with Wide Research: 500-900 credits
• A complex multi-step automation: 1,000+ credits

Monthly plan credits and daily refresh credits do not roll over. Only purchased add-on credit packs never expire — and only while your paid subscription stays active (Fello AI). There is no hard “stop at X credits” budget cap, which is why power users sometimes exhaust a $20 plan in a weekend.

Models powering Manus

Manus is an orchestration layer, not a model lab. It sits on top of third-party frontier models and decides which to use for each phase of a task — reasoning, browsing, coding, writing.

• Anthropic Claude provides the core reasoning and planning. Manus’s reliance on US models is the reason the product was built for markets outside China in the first place (Wikipedia).
• Alibaba Qwen, fine-tuned, handles specialised sub-tasks. A planned Chinese version built with the Qwen team was shelved after Manus relocated to Singapore.
• Multi-model dynamic invocation means a single task can route different steps to different models to use each one’s strengths (MIT Technology Review).

Layered on top are Manus’s three agent tiers, which is what you actually select per task:

Manus has not published which exact Claude or Qwen versions power each tier, and it changes as it upgrades the backbone — so the precise current model mix is data not available from a primary source.

Features deep dive

Agent Mode — the core of Manus

Agent Mode is the product. Give Manus a goal and it works autonomously inside a cloud workspace: it plans the steps, drives a browser, runs code, reads and writes files, calls APIs, and assembles a finished deliverable. Tasks run server-side, so you can close the tab and come back; you can also run up to 20 in parallel on paid plans.

Chat Mode

For quick questions that do not need the full agent, Chat Mode answers directly at a fraction of the credit cost — Manus’s own guidance is to use it for simple queries to avoid burning agent credits.

Wide Research

Wide Research spins up many sub-agents in parallel to cover a topic in breadth and depth at once — dozens of subtasks running together rather than one long sequential job. Since Manus 1.6, all Wide Research sub-agents run on the 1.6 Max architecture, so parallel research inherits the flagship’s accuracy (Manus).

Website and app building

Manus 1.5 (October 2025) added a builder that produces full-stack apps from a single prompt — backend, login, databases, AI features, notifications and analytics. Manus 1.6 extended this to mobile development: describe an iOS or Android app and Manus handles the end-to-end build, a first for the product (Manus).

Design View

Design View, introduced in 1.6, is an interactive image canvas: point-and-click local edits, in-image text rendering, and compositing multiple images into one design. It gives Manus image generation and editing without a separate tool, though it lands at the higher end of credit cost.

Browser operator, Mail and integrations

Manus ships a browser operator for web automation, Mail Manus for email-driven tasks, a Slack integration, AI Slides, an AI image generator and an AI music generator. There is also a developer API and desktop apps for macOS and Windows alongside the iOS and Android apps.

Scheduled tasks

Tasks can be set to run on a cadence — daily, weekly — so recurring work like a morning research digest or a weekly report runs without you re-launching it. Free includes 2 scheduled tasks; paid plans allow 20.

Strengths

It actually completes multi-step work. Manus’s pitch is execution, not conversation. For research, data analysis, slide decks, spreadsheets and app builds, it returns a finished artefact rather than a to-do list — the thing flat-rate chatbots still struggle to do end-to-end.

Manus 1.6 Max raised the success rate. In Manus’s own double-blind testing, 1.6 Max lifted user satisfaction by over 19.2% versus the prior version, with the biggest gains on complex, multi-step tasks and notably strong spreadsheet and web-development performance (Manus). These are vendor-run internal benchmarks, so treat them as directional.

Genuinely broad output. Few rivals span autonomous research, full-stack web and mobile app building, slides, spreadsheets and image editing in one product.

A free tier you can test on. 300 daily credits plus a 1,000-credit starter pack is enough to evaluate real workflows before paying — more generous than most agent products.

Model-agnostic by design. Because Manus orchestrates third-party models rather than training its own, it can adopt the strongest available model for each task without being tied to one lab’s roadmap.

Limitations

The credit system is unpredictable. This is the dominant complaint. There is no fixed per-task price, no hard budget cap, and the in-app estimate is a guide rather than a guarantee — users routinely report a task they expected to cost a few hundred credits clearing 900+ (Fello AI). For variable workflows, monthly spend is hard to forecast.

Credits do not roll over. Unused monthly and daily credits expire; only purchased add-on packs persist, and only while you keep paying.

Ownership is unresolved. Meta’s acquisition was blocked by China’s regulator in April 2026 and remains unsettled, which is a real consideration before committing to an annual plan.

No model of its own. Manus’s capability ceiling is set by the third-party models it orchestrates, so it inherits their limits and any access disruptions.

No real-time voice. Manus is built around autonomous task execution, not conversational voice, where ChatGPT and Gemini lead.

Data and residency questions. Tasks run on Manus’s cloud VMs and the company’s China origins plus unsettled ownership make it a harder sell where data-residency guarantees are mandatory.

Who should use Manus?

Great for:

• Researchers and analysts who want a finished, sourced report from one brief, not a chat thread
• Founders and indie builders who want a working website, web app or mobile app from a prompt
• Operators and finance teams leaning on its strong spreadsheet, modelling and data-analysis work
• Anyone automating recurring work via scheduled agent tasks
• People evaluating autonomous agents — the free tier is a real test bed

Look elsewhere if:

• You need predictable monthly spend — a flat-rate tool like ChatGPT Plus or Claude Pro at $20 is easier to budget
• You mostly ask quick questions — you would be paying agent overhead for chatbot work
• You need voice-first interaction — ChatGPT and Gemini are better
• You require strict data-residency or a settled corporate owner

Historical timeline

Privacy and data handling

Manus is a cloud-first agent: your tasks execute on Manus’s own virtual machines, which read uploaded files, browse the web and run code on your behalf. The company publishes a trust centre and offers SSO and access controls on Team plans, but several specifics that enterprise buyers care about — data-retention windows, training-use policy, regional hosting — are data not available from a clear primary source at the time of writing, so confirm directly with Manus before sensitive use.

Two structural points matter more than the usual fine print. First, Manus relies on third-party models, so your task data also transits whichever provider (Anthropic, Alibaba) handles a given step. Second, the unresolved ownership — Butterfly Effect’s China origins, its Singapore relocation, and the blocked Meta deal — means the legal entity controlling your data could change. For regulated or data-sovereignty-sensitive work, weigh that explicitly.

Manus vs alternatives

Manus competes with the autonomous-agent features now bundled into the big assistants, and with other dedicated agent start-ups such as Genspark.

vs ChatGPT: ChatGPT’s Agent Mode and Operator bring task execution to a $20 flat plan on the largest ecosystem, running GPT-5.5. Manus goes deeper on long, autonomous, multi-tool jobs and on building deployable apps, but you pay per credit rather than a fixed fee.

vs Claude: Claude’s Cowork is a desktop agent that works on your local files and apps, powered by Opus 4.8 — currently the strongest model you can actually use. Manus runs in the cloud and is more of an all-in-one builder. Notably, Claude is one of the models inside Manus.

vs Genspark: Genspark is the closest like-for-like — another dedicated, credit-based general agent orchestrating third-party models. The two trade blows on autonomous tasks; pick on credit value and the specific workflows each handles best.

Getting started

1. Go to manus.im and sign up — the Free plan needs no card and includes 300 daily plus 1,000 starter credits.
2. Pick Chat Mode for quick questions or Agent Mode for autonomous tasks; choose your agent tier (1.6 Lite on Free; 1.6 / 1.6 Max on paid).
3. Describe an outcome, not steps — “Research the top five EV makers by 2025 deliveries and build a comparison slide deck” — and let it run.
4. Watch the credit estimate before launching, and use the dashboard’s usage history to learn your real burn rate.
5. Install the desktop (macOS/Windows) or mobile (iOS/Android) apps to launch tasks from anywhere.

Tip: start on the $40 Pro tier to use the only 7-day free trial Manus offers, learn your credit consumption, then right-size to $20, $40 or $200.

Frequently asked questions

What is Manus AI?

Manus is an autonomous, general-purpose AI agent from the Singapore-based company Butterfly Effect. Unlike a chatbot, it executes multi-step tasks on its own — browsing, coding, writing and file work in a cloud sandbox — and returns a finished deliverable such as a report, slide deck, spreadsheet or working app.

How much does Manus cost?

Manus uses credit-based pricing: Free is $0 (300 daily refresh credits plus a 1,000 starter pack), and Pro is $20, $40 or $200 per month for 4,000, 8,000 or 40,000 monthly credits respectively. Team starts at $20 per seat per month with a 2-seat minimum. Annual billing saves 17%.

Is Manus free to use?

Yes. The Free plan costs $0 and includes 300 daily refresh credits, a one-time 1,000-credit starter pack, one concurrent task, two scheduled tasks, and Manus 1.6 Lite in Agent Mode. It is enough to evaluate the product but not to run heavy or production work.

What models does Manus use?

Manus does not train its own model. It orchestrates third-party frontier models — principally Anthropic’s Claude and fine-tuned Alibaba Qwen — and routes each step of a task to the best one (multi-model dynamic invocation). “Manus 1.6 Lite / 1.6 / 1.6 Max” are agent tiers, not foundation models.

Do Manus credits roll over?

No. Monthly plan credits and the 300 daily refresh credits reset and do not accumulate. Only purchased add-on credit packs persist between months, and only while your paid subscription stays active.

Is Manus owned by Meta?

Not settled. Meta announced an acquisition reportedly worth more than $2 billion in December 2025, but China’s regulator blocked the deal on 27 April 2026, and the ownership question is unresolved as of June 2026. The product continues to operate normally.

Is Manus better than ChatGPT?

They solve different problems. Manus is a dedicated agent that runs long, autonomous, multi-tool jobs and builds deployable apps, billed per credit. ChatGPT is a broader assistant with agent features (Agent Mode, Operator) on a $20 flat plan and a far larger ecosystem. Choose Manus for deep autonomous execution; ChatGPT for predictable cost and all-round use.

What can Manus build?

From a single prompt, Manus can produce research reports, slide decks, spreadsheets and data analyses, full-stack websites and web apps, and — since version 1.6 — iOS and Android mobile apps, plus images via Design View.

Verdict

Manus is the most complete dedicated AI agent you can buy in 2026, and the clearest example of a product that competes on execution rather than conversation. When it works, it turns a one-line brief into a finished report, deck or working app, and Manus 1.6 Max measurably improved its success rate on the hard, multi-step jobs that matter most.

The friction is the credit model. Spend is genuinely hard to predict, there is no budget cap, and unused credits vanish each month — so for variable workflows a flat-rate alternative like ChatGPT or Claude at $20 can be the calmer choice. Layered on top is unresolved ownership after China blocked Meta’s acquisition.

Free ($0): the right way to test whether the agent fits your work before paying.

Pro ($40 with the 7-day trial): the sensible entry point — a real week to learn your credit burn before settling on $20, $40 or $200.

Pro ($200) / Team: for heavy, near-daily agent use or small teams that need shared credits and SSO.

If your work is autonomous and repeatable and you can live with the credit system, Manus is the cleanest paid option in the category. If you mainly want a powerful model to talk to, a flat-rate assistant will serve you better. See the Butterfly Effect provider page for the wider context.

Official links