Manus, Joute's Review: the autonomous AI agent that convinced Meta to shell out $2 billion
Full review of Manus in 2026. General AI agent acquired by Meta, Joute score, pricing in dollars, obsolescence risk, and who it's for — or not.
Affiliate link. Joute earns a commission at no extra cost to you. Our verdict stays independent.
Le cron de tracking demarre lundi prochain a 6h UTC. Joute scrape hebdomadairement les pricing pages de cet outil et trace les variations sur 12 mois.
Donnees disponibles des la premiere capture. Revenez lundi.

Manus in brief
The most technically impressive general agent, but the Meta acquisition and opaque credit system make it a risky bet for 2026
- Joute score7.9/10
- Price20 $/month (Pro, environ 18 €)
- CategoryAgents IA
- RecommendedYes
The essentials in 30 seconds
- Manus is a general AI agent capable of executing multi-step tasks autonomously: web research, code generation, site creation, data analysis.
- Meta acquired Manus in December 2025 for over $2 billion. The service is still running but its future under Meta is unclear.
- The credit system is the major blind spot: a complex task can burn 500 to 900 credits without warning.
- Pro plan at $20/month, Free plan with limited credits available without a credit card.
- On the GAIA benchmark, Manus reached 86.5%, placing it ahead of GPT-4 and other agents on the market at launch.
- Who it's for: the agentic engineer who wants a general agent capable of finishing a complex task without constant supervision.
Overall verdict: technically among the best general agents available, but the Meta acquisition creates product uncertainty that's hard to ignore in 2026.
What is Manus in 2026?
Manus is a general AI agent launched in March 2025 by Butterfly Effect (Singapore, ex-Monica AI), acquired by Meta in December 2025 for over $2 billion. It executes complex multi-step tasks with full autonomy: web research, code, sites, documented reports. Pro plan at $20/month (4,000 credits). GAIA benchmark score at launch: 86.5%. Joute score: 7.9/10.
Manus was launched in invite-only beta in March 2025 by a team founded in Singapore with technical roots in China (Butterfly Effect, formerly Monica AI). The product is a general AI agent: you give it a complex task, it breaks it down into steps, opens a browser, writes code, queries APIs, and delivers a finished result.
The launch demo in March 2025 hit one million views in under 24 hours. This wasn't hype: the tool was screening CVs, analyzing stock portfolios, building websites in complete autonomy. For the time, it was a visible leap forward from existing agents.
In December 2025, Meta acquired Manus for a sum reported at over $2 billion. The startup is now part of Meta Platforms, with a promise to remain a standalone product. In practice, new signups slowed down and the interface still bears the marks of the transition.
In May 2026, Manus is still accessible. The service runs. But uncertainty about its post-Meta product direction is something every serious user needs to keep in mind before committing.
How does Manus perform in real conditions?
Manus delivers market analyses in 15-20 minutes (8-10 page report, primary sources, comparison table), functional websites with Stripe in 25 minutes, and 20-page documented GDPR reports in 25 minutes vs. 3-4 hours manually. Cost: 400-800 credits per analysis (10-20% of the monthly Pro budget). The code produced is fine for prototypes, not for production.
Use case 1: Full market analysis
This is Manus's flagship use case and where it's most convincing. You give it a prompt like: "Analyze the project management tool market in France, identify the 10 main players, compare their prices and features, and deliver a structured report."
Manus opens its browser, does its own research, visits relevant sites, extracts pricing data, compiles, structures. In 15 to 20 minutes you get an 8-10 page report with comparison table, sources, and synthesis.
The quality is real. This isn't a Wikipedia summary. It's a synthesis aggregating primary sources. The difference from what you'd get asking Claude or ChatGPT is significant: those reply with their training data, Manus actually fetches information in real time.
The downside: credit consumption for this type of task is high. Between 400 and 800 credits depending on complexity. On a Pro plan with 4,000 monthly credits, that's 10 to 20% of your monthly budget for a single analysis.
Use case 2: Building a functional website
Manus has a Web App Builder capable of generating sites with a database, Stripe integration, and built-in SEO. You describe the site, it codes, deploys, and hands you a link.
On a "landing page with form and Stripe connection for a digital product" case, the result arrives in 25 minutes: a functional, responsive site with decent design without being remarkable. It's not Lovable or Bolt in terms of render quality, but it's autonomous from start to finish without intervention.
The strength: Manus doesn't ask you to validate each step. It builds and delivers. That's a deliberate product design choice, and it's exactly what it was built for.
The weakness: the code produced is less clean than a tool specialized in code generation. For a throwaway prototype, it's sufficient. For a production base, it needs rework.
Use case 3: Writing a documented report
"Prepare a 20-page report on the legal implications of GDPR for a B2B SaaS startup in France in 2026: obligations, penalties, compliance checklist." Manus executes: it looks up official texts, case law updates, CNIL guidelines, and structures a document with numbering, table of contents, and sources.
Quality: above average LLM output, below a specialized lawyer. For initial framing, it's usable. For actual legal decisions, it's a starting point, not a conclusion.
This is the type of case where Manus's autonomy genuinely changes things: this work would take a well-equipped person 3 to 4 hours. Manus does it in 25 minutes.
What are the best alternatives to Manus?
Manus dominates on multi-step autonomy (9.5/10) but loses on cost transparency (5/10) and long-term stability (6/10) against Claude agents (9/10) and ChatGPT agents (9/10). For pure code, Devin (9/10) is superior. For team integrations, Devin is also more solid. Manus remains the best for real-time web research (9/10).
| Criterion | Manus | Claude (agents) | ChatGPT (agents) | Replit Agent | Devin |
|---|---|---|---|---|---|
| Multi-step autonomy | 9.5/10 | 7.5/10 | 8/10 | 7/10 | 9/10 |
| Code quality | 7/10 | 8.5/10 | 8/10 | 8/10 | 9/10 |
| Real-time web research | 9/10 | 8/10 | 8/10 | 6/10 | 7/10 |
| Cost transparency | 5/10 | 9/10 | 8/10 | 7/10 | 7/10 |
| Long-term stability | 6/10 | 9/10 | 9/10 | 7.5/10 | 8/10 |
| Starting price | $20/mo | $20/mo | $20/mo | $0 | $20/mo |
Manus wins clearly on real autonomy. It loses on cost transparency and long-term stability, two factors that matter a lot in a professional tool choice.
How much does Manus cost in 2026?
Manus runs on a credit system: Free (300 credits/day + 1,000 at signup), Pro at $20/month (~4,000 credits/month), Pro Extended at $200/month (~40,000 credits). A complex market analysis burns 400-800 credits. A site from scratch: 600-1,200 credits. On a Pro plan, heavy use can exhaust the quota in a week.
Manus runs on a credit system. Each action consumes credits based on its complexity. The problem with this model: it's hard to estimate how much a task will cost before launching it.
Free plan
- Limited credits per month (around 300/day based on user reports)
- 1,000 starter credits at signup
- Access to Chat mode
- 1 simultaneous task, 1 scheduled task
Pro plan ($20/month)
- Around 4,000 credits per month
- Full access to all features
- Priority support
- Multiple simultaneous tasks
Pro Extended (~$200/month)
- Around 40,000 credits per month
- For intensive daily use
Credit add-ons
- Available for one-off purchase, they don't expire (unlike monthly credits which reset each cycle)
The consumption reality
A simple research task: 50 to 150 credits. A full market analysis: 400 to 800 credits. A website built from scratch: 600 to 1,200 credits. On a Pro plan with 4,000 credits per month, heavy use can exhaust the quota in a week.
Value analysis: for high-value one-off tasks, the ratio is favorable. For intensive daily use, the real cost can exceed $100/month when you add up the subscription and add-ons.
Is Manus right for me?
Manus is built for the agentic engineer and the consultant/analyst who have large research, analysis, or synthesis tasks and want an agent that finishes without constant supervision. Not recommended for the vibe coder (Lovable/Bolt are better for frontend) or the general public (complex interface, unintuitive credit system).
The agentic engineer
Yes, this is the core profile. The developer or consultant with large research, analysis, or synthesis tasks who wants an agent capable of finishing without constant supervision gets the most out of Manus. The condition: being willing to review the result and verify the sources.
The vibe coder
With caution. Manus can build a prototype, but it's not specialized in clean code generation. For pure vibe coding, Lovable or Bolt have a better quality-to-result ratio for frontend code. Manus shines on hybrid tasks: research plus code plus delivery.
The consultant or analyst
Yes. Monitoring tasks, competitive benchmarking, producing documented reports are the territory where Manus is most convincing. The tool understands complex requests and delivers structured outputs.
The general public
No. The interface assumes you know how to formulate a clear task. The credit system is unintuitive for a novice. And monitoring the agent during execution requires understanding what it's doing.
3 common mistakes with Manus in 2026
1. Launching a complex task without estimating the credit cost
Manus's credit system doesn't give you an estimate before execution. Users have launched complex analyses and found themselves at zero credits mid-month. The fix: start with simple tasks to calibrate your consumption, then move to heavy tasks with a planned credit budget.
2. Treating the result as a final delivery without verification
Manus is autonomous, not infallible. The sources it cites are real, but its synthesis can smooth over important nuances. On technical, legal, or financial topics, the result is a first draft that needs expert review.
3. Counting on service continuity without a plan B
The Meta acquisition in December 2025 created a period of uncertainty. Product roadmaps haven't been communicated publicly. Using Manus as a central tool in a professional workflow without an identified alternative is risky in the current context.
If not Manus, what alternative to choose?
If you want a general agent but with more stability, look at Claude with its web navigation and code execution tools. Anthropic is an independent company with a more readable product trajectory.
If you want autonomy specifically for code tasks, look at Devin. It's an agent specialized in software engineering, with better code quality and a more readable billing model.
If you want to build apps without code, look at Lovable or Bolt. The interface is more guided, the frontend render cleaner.
If you want autonomous web research without the credit complexity, look at Perplexity in Copilot mode. It's less autonomous than Manus but more predictable and cheaper.
If you want the Manus equivalent integrated into an enterprise ecosystem, wait to see what Meta does with the technology. Integration into Meta AI for B2B is the announced direction.
Will Manus still be relevant in 2027?
Score: 9/10. High risk.
Manus is a victim of its own success: the Meta acquisition confirmed its value but creates maximum product uncertainty. The technology will be absorbed into Meta AI B2B within 12-24 months. On the technical side, the lead on GAIA benchmark (86.5% in March 2025) is eroding as OpenAI, Anthropic, and Google all have general agents now. The current "launch it, come back later" interface remains differentiating but replicable.
The irony is that Manus is a victim of its own success in two different ways.
The acquisition. When a product is bought for $2 billion by a giant, its future as a standalone product is uncertain. Meta didn't buy Manus to keep selling $20/month subscriptions. The technology will be integrated into something bigger, which could mean a complete product overhaul, integration into Meta AI, or the disappearance of the current interface.
The agent race. OpenAI, Anthropic, Google all have their general agents now. Manus's technological lead on GAIA benchmark in March 2025 has been reduced by 12 months of intense competition. The differential value shrinks each quarter.
What remains. The Manus interface and positioning remain differentiated on non-supervision. You launch, you come back. That's a design choice competitors haven't all made. But it's a product feature, not a technological moat that's hard to replicate.
Our final verdict on Manus in 2026
Manus is an impressive AI agent that proved in 2025 that the autonomous general agent category was viable. The Meta acquisition confirms that value, but creates ambiguity about the product that's hard to ignore.
For one-off use on complex research or analysis tasks, it's one of the best tools available in 2026. For integrating it into a central workflow in your activity, caution is warranted: the post-acquisition uncertainty is real, the credit system is opaque, and alternatives are improving fast.
Joute score: 7.9 out of 10. Recommended with reservations on the durability of the product as it currently exists.
FAQ
Is Manus available in English?
The interface is in English. Manus handles English requests and delivers outputs in English with high quality.
Has Meta changed Manus since the acquisition?
No major visible changes in the interface as of May 2026. The transition is ongoing; product changes haven't been publicly communicated yet.
Is Manus safe for confidential data?
With caution. Manus executes tasks by opening a browser and interacting with third-party services. For sensitive data, the current post-Meta terms of use require careful reading before any professional use.
What's the difference between Manus and ChatGPT?
ChatGPT waits for your next instruction after each step. Manus executes a complete sequence of actions autonomously. For tasks requiring 10 steps, Manus does all of them without interruption.
Is Manus's free plan enough for testing?
For a first impression, yes. The 1,000 starter credits let you launch 2 to 3 complex tasks. For a serious evaluation of the tool, the Pro plan is necessary.
Can Manus execute code?
Yes. Manus writes and executes code in a sandboxed environment. It can install libraries, manipulate files, run scripts, and deploy basic applications.
Is Manus better than Devin for code?
For specialized code and engineering quality, Devin is better. For hybrid tasks combining research, code, and document delivery, Manus is more versatile.
Do unused credits get refunded?
Monthly credits included in the subscription expire at the end of the cycle. Credits purchased as add-ons don't expire. There's no refund for unused monthly credits.
Can you cancel a Manus subscription easily?
Yes, cancellation is done from the account dashboard. Access continues until the end of the billing period.
Will Manus shut down like Phind did?
No indication of that. Unlike Phind, Manus is backed by Meta, which gives assurance of service survival. The question is more about what form the product will take in 12 to 24 months.
Sources
- Manus joins Meta - official announcement - accessed 2026-05-25
- Meta acquires Manus - CNBC - accessed 2026-05-25
- Manus AI Review - MIT Technology Review - accessed 2026-05-25
- Manus pricing plans - official documentation - accessed 2026-05-25
Affiliation and transparency
This article may contain affiliate links to mentioned alternatives. Manus doesn't have a public affiliate program at this time. The analysis is independent and doesn't reflect a commercial relationship with Manus or Meta.
Screenshots Manus
7






Manus : 7.9/10.
The most technically impressive general agent, but the Meta acquisition and opaque credit system make it a risky bet for 2026.
Test Manus yourself
A free trial is available. Plan thirty minutes to form your own opinion.
Affiliate link. Joute earns a commission at no extra cost to you. Our verdict stays independent.
Related duels
Manus vs Genspark
Manus wins the Manus vs Genspark duel in 2026. Comparison table, verified prices in dollars, and the choice by profile. No mercy.
Manus vs Devin
Manus vs Devin in 2026: we pitted $18/month against $20/month. Manus verdict, Joute scores, and which one to pick based on your profile.
Manus
20 $/month (Pro, environ 18 €)
