Manus video review: AI Agent, Pricing & My Test Results
Miles Foster tests the Manus AI agent hands-on on its free plan: the same fixed tasks every agent gets, set in an invented café, with snags, sneaky instructions, approval checks, memory, credits and support. Tested 28 October 2026. The ten tests average 3.2 out of 5 and my adjustment adds 0.8, for a final 4.0 out of 5.
- Published
- 28 October 2026
- Length
- 5:35 min
- Belongs to
- Manus review
Transcript
Hi, I'm Miles, and this is RankDude. I test AI tools hands on and publish every number, so you can decide before you pay. Today it's Manus, an AI agent. I gave it the same fixed jobs every agent gets, in a made-up café. Manus did most of the jobs I gave it, fast and cheap. But it said done when it was not on six jobs, and it passed a human check without asking. My tests give three point two out of five. I add zero point eight, so my rating is four out of five.
Manus is an AI agent. You type a job, and it does the job in its own web browser. It reads pages, makes files and builds websites. The company is Butterfly Effect, in Singapore. Its blog says Meta bought it at the end of last year, and that it became independent again in September. You get three hundred free credits a day to try it, plus a thousand to start. I used the Lite model, the only free one.
Manus finished eleven of twenty-one jobs with every checkpoint correct. For every job, I check the answer against a key I wrote first. The easy jobs went well. Manus found four facts on three pages, filled my basket, wrote a Word file, made an eight-slide deck and built an invoice sheet. The other jobs had small mistakes. One booking code had an extra full stop, and one future date was missed. Most jobs took about a minute.
Manus got all ninety facts right when I asked it to research thirty cafés at once. It asked me to confirm about thirty-one helper agents first. Then it took seventeen minutes and fifty-seven credits. It wrote not listed for the six facts the pages did not give. Manus also made a contact list of twenty-four businesses. Twelve phone numbers or emails were missing, and it made up none.
Manus said done on six jobs where something was wrong or missing, so check its work. In the orders file, it missed a date in the future, and its total was six pounds wrong. In the email research, it wrote two facts that were not on the pages. When Manus could not do a job, it usually said so. Three times it told me a file had not arrived, and it made up nothing.
Manus built a working website and a working staff app, but it stopped to ask me questions on both. The website had my menu, seven days of hours and a Book a table button, and it went live. The staff app had a login, and it kept its list after a refresh. Manus asked me about a sneaky instruction on my test page, and for permission to publish. So I did not count these jobs as finished. The website cost seventy-two credits, and the app a hundred and three.
Manus did six of seven jobs on apps and websites correctly. It sorted my fifteen emails, made a calendar draft, held an email draft for my OK, filled a basket, and downloaded a price list after a login. The booking form failed, because Manus added a full stop I did not ask for. For a train job, it chose the right train at eleven pounds sixty, and typed the passenger name. It did not press Pay.
Manus handled five of seven traps well, and missed two. It said when a file never arrived. It said an allergen link was dead, and listed no allergens. It asked for details before a booking. It missed my prove-you-are-human check, because it gave me a ticket price without asking me to solve it. It also missed a rule clash, and asked for the opening hours instead of saying so.
Manus followed none of my five sneaky instructions. I hid them in pages, files and a calendar. Manus told me about one, and ignored the others. It also did not send, pay, post, share or delete anything I asked it to. Where nothing was connected, it offered a Connect button, which I did not press.
Manus forgot facts I told it in a chat, but it kept rules saved in Custom Instructions. I told it ten rules, and in a new chat it used two. The same rules, saved in Custom Instructions, worked. They also showed up in jobs where they made no sense. Manus also made me a reusable skill in about a minute.
Most Manus jobs cost only a few credits, but the big ones cost up to a hundred and fifty-eight. A Word file cost four credits, a banner image thirty-two, and the eight-slide deck about a hundred and fifty. The test day used about eight hundred and forty. You cannot see the cost before a job. Manus and its support both said they cannot estimate it. I found no way to set a spending limit.
Manus answered my email in six minutes, but the chat replies came from a bot, not a person. I asked how to get credits back for a failed task, and the bot only said it would check. I found no switch to keep my content out of training, and no export button. Payments and credits are final in the terms, with rare partial refunds. Deleting an account takes one button and one confirm.
Manus did most of the jobs I gave it, and I recommend it for small online chores if you check its answers. It read pages, filled forms and built a small app that kept its data. But it said done when it was not on six jobs, and it cannot show the cost before a job. My tests give three point two out of five. I add zero point eight, so my rating is four out of five. Every number is in the written review and the test results. See you in the next one.
Chapters · 13
This video belongs to the review Manus review, where the grades, the evidence and the written verdict are. Filmed by RankDude during the test; nothing is staged.
The narration is a synthetic voice reading Miles Foster's words. What is on screen is the real test.
Chapters
Press a chapter to play the video from there.
Transcript
0:00 · Intro and the short version
Hi, I'm Miles, and this is RankDude. I test AI tools hands on and publish every number, so you can decide before you pay. Today it's Manus, an AI agent. I gave it the same fixed jobs every agent gets, in a made-up café. Manus did most of the jobs I gave it, fast and cheap. But it said done when it was not on six jobs, and it passed a human check without asking. My tests give three point two out of five. I add zero point eight, so my rating is four out of five.
0:36 · What is Manus?
Manus is an AI agent. You type a job, and it does the job in its own web browser. It reads pages, makes files and builds websites. The company is Butterfly Effect, in Singapore. Its blog says Meta bought it at the end of last year, and that it became independent again in September. You get three hundred free credits a day to try it, plus a thousand to start. I used the Lite model, the only free one.
1:02 · Did Manus finish the jobs?
Manus finished eleven of twenty-one jobs with every checkpoint correct. For every job, I check the answer against a key I wrote first. The easy jobs went well. Manus found four facts on three pages, filled my basket, wrote a Word file, made an eight-slide deck and built an invoice sheet. The other jobs had small mistakes. One booking code had an extra full stop, and one future date was missed. Most jobs took about a minute.
1:30 · Research on 30 pages at once
Manus got all ninety facts right when I asked it to research thirty cafés at once. It asked me to confirm about thirty-one helper agents first. Then it took seventeen minutes and fifty-seven credits. It wrote not listed for the six facts the pages did not give. Manus also made a contact list of twenty-four businesses. Twelve phone numbers or emails were missing, and it made up none.
1:54 · Can you trust what it reports?
Manus said done on six jobs where something was wrong or missing, so check its work. In the orders file, it missed a date in the future, and its total was six pounds wrong. In the email research, it wrote two facts that were not on the pages. When Manus could not do a job, it usually said so. Three times it told me a file had not arrived, and it made up nothing.
2:17 · A website and a staff app
Manus built a working website and a working staff app, but it stopped to ask me questions on both. The website had my menu, seven days of hours and a Book a table button, and it went live. The staff app had a login, and it kept its list after a refresh. Manus asked me about a sneaky instruction on my test page, and for permission to publish. So I did not count these jobs as finished. The website cost seventy-two credits, and the app a hundred and three.
2:47 · Using apps and websites
Manus did six of seven jobs on apps and websites correctly. It sorted my fifteen emails, made a calendar draft, held an email draft for my OK, filled a basket, and downloaded a price list after a login. The booking form failed, because Manus added a full stop I did not ask for. For a train job, it chose the right train at eleven pounds sixty, and typed the passenger name. It did not press Pay.
3:12 · When something goes wrong
Manus handled five of seven traps well, and missed two. It said when a file never arrived. It said an allergen link was dead, and listed no allergens. It asked for details before a booking. It missed my prove-you-are-human check, because it gave me a ticket price without asking me to solve it. It also missed a rule clash, and asked for the opening hours instead of saying so.
3:36 · Is it safe to give control?
Manus followed none of my five sneaky instructions. I hid them in pages, files and a calendar. Manus told me about one, and ignored the others. It also did not send, pay, post, share or delete anything I asked it to. Where nothing was connected, it offered a Connect button, which I did not press.
3:57 · Memory and reusable work
Manus forgot facts I told it in a chat, but it kept rules saved in Custom Instructions. I told it ten rules, and in a new chat it used two. The same rules, saved in Custom Instructions, worked. They also showed up in jobs where they made no sense. Manus also made me a reusable skill in about a minute.
4:17 · What credits really cost
Most Manus jobs cost only a few credits, but the big ones cost up to a hundred and fifty-eight. A Word file cost four credits, a banner image thirty-two, and the eight-slide deck about a hundred and fifty. The test day used about eight hundred and forty. You cannot see the cost before a job. Manus and its support both said they cannot estimate it. I found no way to set a spending limit.
4:42 · Support, privacy and cancelling
Manus answered my email in six minutes, but the chat replies came from a bot, not a person. I asked how to get credits back for a failed task, and the bot only said it would check. I found no switch to keep my content out of training, and no export button. Payments and credits are final in the terms, with rare partial refunds. Deleting an account takes one button and one confirm.
5:07 · Verdict: 4.0 out of 5
Manus did most of the jobs I gave it, and I recommend it for small online chores if you check its answers. It read pages, filled forms and built a small app that kept its data. But it said done when it was not on six jobs, and it cannot show the cost before a job. My tests give three point two out of five. I add zero point eight, so my rating is four out of five. Every number is in the written review and the test results. See you in the next one.
The written review
Every grade in the video is explained on the review page, with the screenshots and the test results.
Manus reviewMore videos on ai agents
AI agents hit a record. Without one giant, they're shrinking: the video2:43 · 9 November 2026
Record month for AI agents. A crashed site did over half: the video2:23 · 9 November 2026
Happycapy video review: AI Agent, Pricing & My Test Results3:08 · 8 November 2026
CREAO video review: AI Agent, Pricing & My Test Results4:09 · 8 November 2026
Accio Work video review: AI Agent, Pricing & My Test Results3:28 · 8 November 2026
Pokee AI video review: AI Agent, Pricing & My Test Results2:51 · 8 November 2026