What Manus Is Actually For, Now Meta Doesn't Own It

On September 1, 2026, Manus announced that it had "formally resumed independent operations," with its founding team continuing to lead the company. That one line from the company's own blog replaces the story this guide told in August, when Manus was a startup whose founders were still trying to buy themselves back from Meta.

If you are deciding whether to pay for Manus today, the ownership drama is mostly behind you and the product has moved on. Four weeks after the independence post, Manus shipped version 2.0, then a bring-your-own-key option called Flex. This update rewrites the guide around what the product is for now, what it costs, and who should skip it.

What changed since August

The August version of this guide rested on one fact: Meta had agreed to buy Manus in December 2025, and that arrangement was coming apart. That history still matters for context, and it is the only part of it we keep. We no longer repeat any deal price or valuation, because Manus has not published one and we have not re-sourced the figures from the earlier coverage.

Here is what the company itself has said since, with the date of each post:

Two of those rows deserve a closer look because they change how you would budget. The first is the data note. Manus says some users had to back up and restore their data and lived through a temporary interruption to access. If you used Manus before September and your old projects look empty, the restoration portal and the Data Back Up and Restoration pages in the help center are where the company tells you to start.

The second is Flex, covered in its own section below.

What Manus 2.0 adds to the job description

Manus describes 2.0 as "not a version update" but a new architecture. Four pieces matter for everyday use.

• Cascade is the in-house agent runtime that plans and carries out the work. Manus says that in one tested configuration it used 23.2% fewer tokens, finished tasks in 28.2% less time and cost 32% less to run than the previous system. That is a vendor figure from a single configuration, so treat it as direction, not a promise about your workload.
• Cloud Computer is a dedicated environment you can purchase for work that has to keep running, such as the server behind a multiplayer game or a permanent home for an automation.
• Automations extend scheduled tasks. Work can now start when something happens in a connected service: an incoming email, a calendar entry, a chat message or a change in a Notion page can each start a run.
• Manus Studio is the new home for creative work, and Manus separately announced a video editor and a game builder on October 1. Cue, a personal-agents app, is a separate product and not part of the Manus subscription described here.

The practical shift is from "give it a goal and wait" to "set it up once and let it watch." Anyone who used Manus mainly to produce a single research report will not notice most of this. Anyone who wanted a recurring digest or an email-triggered workflow has a reason to look again, and should compare it against workflow tools like Zapier, Make and n8n, which have done event triggers for years.

What $20 to $200 a month buys

Hokai's record of the Manus plans, last reviewed on September 29, lists these tiers:

Prices move, so check the live listing before you pay. The pattern that has not moved is the unit: Manus meters work in credits, and a complex agent run has been reported to burn 500 to 900 of them with no estimate shown before it starts. Unused monthly credits expire. For a freelance researcher who needs one deep report a week, that is workable. For anyone budgeting per task, it is the main source of surprise.

Flex: the first real change to the cost model

Manus Flex lets you connect your own API key from a supported inference provider. Manus says it keeps providing the agent system that plans the work, uses tools and operates across environments, while the provider bills you directly for model inference. Flex setup includes a model choice and a reasoning-effort setting. Anything else a task touches, such as hosting or other infrastructure, still consumes Manus credits.

Manus's own announcement of Flex, captured 11 Oct 2026.

The launch partners are OpenRouter, Fireworks and Modal. In plain terms:

1. If you already pay a provider for model access, you can point Manus at that relationship instead of burning credits on the same inference.
2. If you want to pick a cheaper or stronger model for a given job, Flex is the first time Manus has allowed that.
3. If you never touch an API key, Flex changes nothing for you, and the credit plans above still apply.

A caution on the savings: Manus has not published a comparison of credit cost against provider cost for the same task, so we cannot say Flex is cheaper. It splits the bill in two, which makes it harder to see the total. Track both for a week before deciding.

Who this affects

Returning users with data stuck in a backup are the group with the most immediate to-do. Teams that held off because of the ownership uncertainty now have a clearer answer on who runs the company, though a self-described independent agent lab is still a startup and a startup's support depth is not a large vendor's.

Developers and agencies are the second group. Flex and Cloud Computer both aim at people who want the agent system without the lock-in to one credit meter. Casual users on the free tier are the third, and for them the main news is that Studio gives the free plan more kinds of output to try.

Where the alternatives still win

Manus is not the right pick for every agent job. Three cases favor something else.

First, if you want one assistant that lives inside tools you already pay for, ChatGPT bundles an agent mode with a chat product and a larger ecosystem. Our Manus vs ChatGPT agent comparison walks through the trade-offs task by task.

Second, if your work is research with citations, a dedicated tool like Perplexity or Genspark can be a better fit than a general agent, and the roundup of the best AI for deep research compares them on that single job.

Third, if you are building agents for a business process rather than using one, a builder such as Lindy or the options in our best AI agent builders list gives you control that a finished product does not. Coding-first work belongs with tools like Replit, Lovable or Devin.

For the full field, start with the best AI agents of 2026, the agentic AI roundup and the business operations list. If you are weighing the newest consumer agents, what OpenAI's dots do shows how the same idea is being packaged elsewhere.

The benchmark caveat still stands

Many reviews of Manus lead with its 86.5% GAIA Level 1 score. That figure is self-reported from the March 2025 launch, and we have not found an independent re-run.

In April 2026 a UC Berkeley team showed that GAIA's validation answers were publicly available, which let an exploit agent score about 98% without solving a task. That does not prove the Manus number was inflated. It does mean the benchmark behind the claim has a documented leak.

The better test is cheaper: run one real task on the free tier and read the output.

What to watch next

Three things will tell you whether this independence holds up and whether the product earns the new tiers.

First, whether Manus publishes numbers for Flex. A side-by-side of credit cost and provider cost for the same task would settle the biggest open question in this guide, and nobody outside the company can produce it cleanly because credit burn varies by run.

Second, whether Automations prove reliable. Event-triggered work is only useful if it fires when it should and stops when it should. A missed email trigger is an annoyance, but a runaway automation that keeps spending credits is a bill. Watch for how Manus handles spending limits on always-on jobs, and test with a small credit allowance before you wire it to a busy inbox.

Third, whether the Meta episode leaves any lingering questions about data handling. The company's own posts describe a backup and restore process for affected users and a temporary access interruption. They do not describe how it was resolved beyond the portal. If you store sensitive client material in Manus, ask support what happened to it during the transition before you rely on it again.

What would change this guide's verdict: an independent benchmark re-run of the agent, a published Flex cost comparison, or a pricing change that moves the credit tiers. Any of those would be worth a re-read of the plans above.

Why the credit meter matters more than the headline price

A plan price tells you the entry fee. The credit meter tells you the running cost, and it is the part most buyers misjudge. Four thousand credits sounds generous until a single complex run takes a sixth of them. If you plan to run several agent jobs a day, the Standard plan may last under a week, and the jump to Customizable or Extended is a decision about volume, not features.

Flex shifts that math for anyone who already has a provider account. The model portion of a run moves to a bill you can see in the provider's dashboard, with its own usage graphs and limits. The Manus credits then pay only for the surrounding services. That is more transparent, and it is also one more invoice to watch.

How to test Manus without spending anything

The free tier is enough to answer most of the questions this guide raises, if you use it on purpose rather than on a toy prompt.

Pick one job you do every week and that takes you at least an hour. Good candidates are a competitor summary, a weekly metrics digest, a slide deck from a rough outline, or a small landing page. Write the brief the way you would for a junior colleague, with the audience, the format and the deadline. Then let the agent run and note three things: how many of the 300 daily credits it used, how much of the output you could ship without editing, and what it got wrong.

Run the same brief once in a chat product like ChatGPT and once in Manus. Compare time to a usable draft rather than the polish of the first answer, because the point of an agent is the hours you get back. If Manus saves you less than an hour a week, the $20 plan will not pay for itself, and the free tier is the right place to stay.

If the result is good and the job repeats, try it as an Automation with a trigger you control, such as a calendar event you create yourself. That tells you more about reliability than any benchmark, and it costs a handful of credits. For ideas on which jobs agents handle well today, our agentic AI roundup groups them by task, and the Lovable and Replit pages show where app-building agents overlap with Studio.

What to do about it

1. Check your account. If you used Manus before September 1, look for unrestored data in the restoration portal.
2. Run one task on the free tier. Pick an Automation if recurring work is your reason to care.
3. If you already pay a model provider, try Flex with a single project, and log both bills.
4. Compare before you commit to $200. The Extended plan only makes sense once you have measured your own credit burn. Where a specialist like ElevenLabs beats a general agent on one job, use the specialist.
5. Still unsure which agent fits your work? Smart Match asks about your task and budget, then returns agents worth trying.

Re-checked October 11, 2026: the pricing tiers above come from Hokai's September 29 record, and the company's own posts are the source for everything dated September 1 through October 1.

Frequently asked questions

Does Meta still own Manus?

No. Manus announced on September 1, 2026 that it had formally resumed independent operations, with its founding team continuing to lead the company. Some users had to back up and restore data during the transition, and Manus says the restoration portal has no deadline.

How much does Manus cost in 2026?

Hokai's September 29 record lists a free tier with 300 daily credits, Pro Standard at $20 a month for 4,000 credits, Pro Customizable at $40 for 8,000, Pro Extended at $200 for 40,000, and a Team plan at $40 per seat with a two-seat minimum. Complex agent tasks have been reported to burn 500 to 900 credits each.

What is Manus Flex?

Flex is a Manus module, launched September 29, 2026, that lets you power the agent with your own API key from a supported provider. The provider bills inference directly, while other services used in a task still consume Manus credits. OpenRouter, Fireworks and Modal are the launch partners.

What did Manus 2.0 add?

Manus 2.0, announced September 28, 2026, added the Cascade agent runtime, purchasable Cloud Computer environments, event-triggered Automations and Manus Studio. Manus says Cascade used 23.2% fewer tokens and cost 32% less to run than its previous system in one tested configuration.

Is Manus's 86.5% GAIA benchmark score reliable?

The score is self-reported and has not been independently re-run since the March 2025 launch. In April 2026 a UC Berkeley team showed GAIA's validation answers were publicly available, which does not prove the Manus figure was inflated but does mean the benchmark has a documented leak.

Covered in this guide

• manus: Autonomous AI agent that plans and executes multi-step work from one prompt: research, documents, apps and video. The September 2026 release adds Cascade, Automations and Studio. Free tier available.
• chatgpt: ChatGPT is OpenAI's AI assistant, with 1.2 billion weekly users by OpenAI's count, the GPT-6 model family and plans that run from a free tier to a premium Pro tier.
• fireworks-ai: Enterprise inference platform for open models: serverless tokens from $0.10 per 1M, managed fine-tuning per training token, and dedicated NVIDIA GPUs from $8.00/hr. $1 free credits, self-serve, no cold starts.
• genspark: Genspark is an AI super agent that runs 8+ LLMs and 80+ tools to build slides, research reports, and apps, with a free daily-credit tier and paid Plus/Pro plans.
• lovable: Build something lovable. Create apps and websites by chatting with AI.
• make: No-code visual automation platform to connect apps and build AI workflows at scale
• modal: AI infrastructure that developers love: serverless compute for ML inference, training, and batch processing
• n8n: AI workflow automation platform for technical teams
• replit: Cloud coding environment where you describe an app in plain English and AI builds and deploys it.

Sources