Manus 2.0: Cascade, Studio, Cue and what is confirmed
The official launch introduces Cascade, Manus Studio, Cloud Computer, Automations and Cue. The reported gains are from one company test.
View the original post ↓Manus describes Cascade, Studio, Cloud Computer, Automations and Cue in its official 2.0 launch. The efficiency percentages come from one vendor-tested configuration.
Manus published Introducing Manus 2.0 on Monday, 28 September 2026. The company frames it as more than a version bump: a new agent harness called Cascade, a rebuilt desktop workspace called Manus Studio, persistent Cloud Computer instances, event-driven Automations, and a standalone personal-agent app called Cue.
This note separates three layers that launches like this usually mix:
- What Manus wrote on its own blog and help pages.
- What it showed on X the same day.
- What a user can actually open today, with which limits.
It is not a hands-on review. The Cascade percentages below are Manus’s own test, not a measurement from this desk.
Where this note started
The first post I opened was the Japanese official account, @ManusAI_JP, which introduced 2.0 and pointed to Studio, Cue and the Japanese blog post.
The global product account posted the same day with a longer launch film:
Both posts are useful as primary media. The text that follows is grounded in the English product blog, with the Japanese post used only where it adds a download or locale URL.
Why 2.0 is not just another weekly agent update
Manus spent 2025 as one of the more visible “general agent” products: a system that is supposed to finish work (research, files, sites, code), not only answer in chat.
On 1 September 2026 the company published Manus Resumes Independent Operations. The founding team said it would keep leading the company as an independent agent lab. Some users had to back up and restore data after a temporary access interruption; users who were not affected did not have to do anything.
That post does not mention Meta. Press coverage of the 2.0 launch, including Bloomberg’s same-day report, treats Manus as a company that is pushing again in the agent market after an acquisition process that did not complete. For this article, the operational fact that matters is the official one: Manus says it is independent again, and 2.0 is the first large product surface after that statement.
What Manus 2.0 actually contains
The official post groups the release in three blocks.
Block / Product / Job it claims to do
- Block: Foundation; Product: Cascade; Job it claims to do: In-house agent harness that starts light and adds tools only when the task needs them
- Block: Foundation; Product: Cloud Computer; Job it claims to do: Paid dedicated environment that stays up when a laptop closes
- Block: Foundation; Product: Automations; Job it claims to do: Workflows that start from events, not only from a clock
- Block: Creation; Product: Manus Studio; Job it claims to do: Desktop workspace; loads only the tools the current project needs
- Block: Creation; Product: Video Editor + Alchemy; Job it claims to do: Timeline editing of agent-made video, plus a higher-ambition generation mode
- Block: Creation; Product: Game Dev; Job it claims to do: Template-to-playable games, with optional multiplayer on Cloud Computer
- Block: Creation; Product: Computer Use / Remote Control; Job it claims to do: Authorized session on your machine, visible while it works
- Block: Personal; Product: Cue; Job it claims to do: Separate app: agents with their own email, phone, wallet and computer
If you only remember one sentence: Manus 2.0 splits “finish a work project” (Studio + Cascade) from “run a personal agent that can message and pay” (Cue). They share infrastructure. They are not the same app.
Cascade: the claim, and the limit of that claim
Cascade is described as the latest iteration of Manus’s in-house agent harness. The design idea is ordinary and still easy to miss: do not drag a video editor, a game engine and a spreadsheet into every task. Start small. Bring the specialist surface in when the work requires it. Keep the brief, the page, the video and the automation inside one project.
The efficiency line is specific, and it is also scoped:
In one tested configuration, Cascade used 23.2% fewer tokens, finished tasks in 28.2% less time, and cost 32% less to run than our previous system.
That sentence is from the official blog. It is one tested configuration, run by Manus, against Manus’s previous system. It is not a public leaderboard, not a third-party audit, and not a promise that your next project will be 32% cheaper.
What you can take from it without over-reading:
- Manus is optimizing the harness (how tools and context are loaded), not only swapping the underlying model.
- Token use, wall-clock time and operating cost moved together in that test. That is the interesting part. A harness that is faster but burns more tokens is a different story.
- Until someone outside Manus publishes a task suite with the same prompts on 1.x and 2.0, treat 23.2 / 28.2 / 32 as vendor test data.
Cloud Computer: persistence, not a free extra
The blog’s examples are concrete. A multiplayer game needs a server that does not die when the laptop lid closes. An automation that watches a mailbox all day needs a machine that is still there at 03:00.
Cloud Computer is that machine. Manus says you purchase it. Thurrott’s same-day write-up also flags it as a dedicated environment at extra cost. Pricing is not in the launch post; confirm it in the product before you plan a workload around it.
This is the piece that turns “the agent made a game” into “friends can actually join.” Without a persistent host, multiplayer is a demo. With one, it is an operational object you are now paying to keep alive.
Automations: from cron to events
Scheduled Tasks already existed: Manus starts work at a time you set. Automations add the other half of a real workflow. A new email, a change in ad performance, a calendar event, a Slack message or a Notion update can start a task.
The promised interface is one prompt: tell Manus what to watch and what to do next. That is the product claim. The implementation detail that matters in practice is permissioning — which mailbox, which ad account, which Notion workspace — and whether the trigger is reliable enough that you would let it spend money or send mail without a human in the loop.
Until you have seen a trigger fire twice on the same event and zero times on a false one, keep Automations in “draft replies / draft reports” rather than “publish and charge.”
Manus Studio: the desktop app stops being a thin wrapper
The desktop app is renamed Manus Studio and described as a shared workspace for people and AI. The same Cascade rule applies: Studio is not supposed to load every tool into every session.
The launch lists two layers of work:
- Everyday: documents, spreadsheets, PDFs, slides.
- Professional: websites, code, games, video.
Video and games get new environments because, in Manus’s words, the bar is higher there.
Video Editor
The example in the blog is a launch ad due tomorrow. Manus has a first cut. You want a warmer song and your own product shot instead of a generated one.
The old loop was: write a new prompt, regenerate the whole video. Studio’s answer is a timeline. Clips, images, text, motion graphics and audio arrive as separate pieces. You swap the song, drop in footage, move an animation, export. You can also hand the timeline back and ask Manus for another pass on top of your edits.
Manus positions Video Editor for 30–60 second product ads, AI UGC, data motion graphics, tutorials, vlogs and motion-led videos, and says you do not need editing experience.
That last clause is marketing. A timeline is still a timeline. The useful change is architectural: the agent’s output becomes an editable project, not a sealed MP4. That is the same idea other creative-AI tools have been racing toward. Whether Studio’s timeline is pleasant is a hands-on question.
Alchemy mode
Alchemy is the “quality matters most” switch. Manus says it takes creative direction, combining video and code generation, with tighter visual control, better motion and better pacing. You can still fine-tune on the timeline after.
Read it as a higher-spend generation profile, not as a new physics engine. If you test it, keep the same brief in the default editor and in Alchemy and compare the timeline, not the trailer Manus cut for X.
Game Dev
Game Dev is aimed at people who have already seen AI produce a “decent” game and want the version they actually meant. It combines video, image and coding models. Max Ultra is named as the quality mode.
The Edit panel is the control surface: watch the game run, manage assets, edit code, change movement, art and scenes by hand. The example is petty on purpose — one villager’s hair from brown to silver — because that is how you tell whether the tool edits a part or rebuilds the world.
Publishing is a website anyone can play from a link. Multiplayer is the part Manus spends the most words on, because that is where amateur projects usually die. The documented path is: Configure → Multiplayer → Select Cloud Computer → Buy one. Manus says it handles hosting and networking for racing, head-to-head, competitive and 3D games.
Again: multiplayer is tied to a purchased Cloud Computer. The “few clicks” sentence does not make the server free.
Computer Use and remote control
This is the other half of “agent that does work.” A normal remote-desktop app still makes you tap every button on a phone screen. Computer Use, in an authorized session, lets Manus work in a visible workspace on your computer with the files, browser and apps you approved.
The cab example: ask it to find the latest deck in a launch folder on the home machine and send it to you, while you watch the desktop live.
The constraint to keep is authorization. If the session can see the browser, it can see more than the deck. Treat Computer Use like giving a contractor a chair in your office, not like a magic folder picker.
Cue: a different product, same plumbing
Cue is not a skin on Studio. It is a standalone app on phone and desktop, built on the same infrastructure, with iOS still waiting on App Store review at launch.
The identity model is the headline:
- each agent gets its own email
- its own phone number
- its own wallet
- its own computer
Manus says the agent can send messages, pay inside a budget you set, finish a task on its own machine, take your calls and leave a summary in Cue.
There is a multi-agent mode: several agents in a group chat with a shared goal, handing work along a chain (research venues → shortlist → draft a deck) while you keep the final call.
And there is an offline-world claim: at a restaurant, scan a QR code and the agent can order or hold a place in line. Bloomberg repeats the food-ordering / hold-a-spot example. That is a product demonstration until you have seen it survive a real queue, a real payment and a real cancellation.
Access on day one
Cue is early access, free with an invite code. The published starter code is MEETCUE, limited, first come first served. People who get in receive extra codes to share.
Official entry points:
- Cue web and desktop: cue.im
- Manus 2.0 web: manus.im
- Desktop Studio: manus.im/desktop
Codes rot. If MEETCUE is dead when you read this, use the live page, not this paragraph.
What to try first, if you already use agents
A launch this wide is a good way to waste an evening clicking every new noun. A tighter pass:
- Open a task you already trust Manus to finish (a short brief, a table, a one-page site). Compare 2.0 against your last run. You are checking whether Cascade feels lighter, not whether the trailer was pretty.
- If you make short ads, generate one 30–60s cut and open it in Video Editor. The test is whether you can swap audio and one shot without regenerating the whole file.
- Leave Game Dev and multiplayer for later unless you are willing to buy Cloud Computer and treat it as an experiment with a bill.
- Cue: create one agent with a tiny budget and a mailbox you do not care about. Do not hand it your primary WhatsApp, bank or ads account on day one.
- Automations: watch one trigger (calendar or Notion) and confirm it fires once.
That sequence matches how I would audit any agent stack for a real workflow: one bounded job, one editable artefact, one permission surface.
What is not confirmed
- That Cascade’s 23.2 / 28.2 / 32 numbers will show up on your workload.
- That Alchemy mode is “better” in any sense other than Manus’s own qualitative description.
- That Cue agents paying and calling will be allowed, stable or legal in every country where you travel.
- That iOS Cue is available. At launch it was not; it was waiting on App Store review.
- Any ranking, revenue or “best agent of 2026” claim. Those belong to other people.
Frequently asked questions
Is Manus 2.0 a new model or a new product line? Both, in Manus’s telling: a new harness (Cascade) plus new surfaces (Studio, Cloud Computer, Automations, Cue). The blog does not publish a model card for a single named base model in this post.
Is Cue included with a Manus subscription? The launch post presents Cue as a standalone early-access app with its own domain and invite code. Confirm current bundling on the pricing page; do not assume it is a free extra forever.
Do I need Cloud Computer to use Studio? No. The documented need is persistence: games that stay up, automations that run overnight. Short desktop sessions do not require it.
Should a business put Cue on a customer-facing number? Not from this article. An agent with its own phone and wallet is a new operational surface (consent, spend limits, transcripts, handover when it fails). That is a policy decision, not a feature checkbox.
Takeaway
Manus 2.0 is a split. Studio is for projects you want to edit after the agent drafts them. Cue is for agents that need an identity in the world. Cascade is the bet that loading fewer tools by default is how you get both. The only number in the launch that looks precise is also the one you should treat as a vendor test.