Launches · September 29, 2026 · 8 min read
Manus 2.0 ships Cascade and Studio on the main app, while Cue gives each agent its own email, phone number, wallet, and computer behind an invite code, and every efficiency percentage on the page comes from one tested configuration.
On September 28, Manus published a blog post titled “Introducing Manus 2.0.” The company opens with a line that the release is not a version update. It calls the ship a new architecture, new products, and new capabilities. Manus 2.0 is on the web, desktop, and mobile now. Cue, a separate app for personal agents, is in early access on the same platforms, with iOS listed as coming after App Store review. The gate for Cue is an invitation code. Manus published one code, MEETCUE, for a limited number of first-come users, and said everyone who gets in receives extra codes to share. An agent with its own phone number and its own wallet stopped being a thought experiment on that day and became a signup flow.
Cascade and the one configuration
Cascade is the name Manus gives its in-house agent harness in 2.0. The blog and the Help Center describe the same design goal. A project stays light at the start. Specialized capabilities arrive only when the work needs them, so a brief, a page, a video, and an automation can remain one connected project instead of four separate stacks eating context.
The efficiency claim is numeric and narrow. In one tested configuration, Cascade used 23.2 percent fewer tokens, finished tasks in 28.2 percent less time, and cost 32 percent less to run than Manus’s previous system, according to the launch post and the Help Center article “What is new in Manus 2.0?” The Help Center repeats the same three percentages and adds an explicit caveat: these results come from one tested configuration, not an overall average. There is no public benchmark name, no task list, and no third-party verification in the materials Manus published. Every efficiency figure in this piece is Manus’s own, from that single described setup.
Studio, Cloud Computer, and Automations
The desktop app is upgraded to Manus Studio, a shared workspace where Manus works with files already on the machine in a connected and authorized session. The Help Center says most projects start in a folder, not in a blank prompt: phone footage, screen recordings, product photos, half-finished sites, notes. Studio’s loop is the same across formats: Manus makes a first version, the user shapes it with direct controls, and either side can take the next pass.
The launch post lists the output types Studio covers in this release: documents, spreadsheets, PDFs, slides, websites, code, games, and video. Two add-ons get their own environments. Video Editor hands back a timeline after the first cut, so a user can swap music, trim a pause, or drop in a real product shot without regenerating the whole piece. Game Dev starts from playable templates and adds an Edit panel for assets, code, and scenes. Multiplayer routes through a purchased Cloud Computer so a game server can stay up when the laptop closes.
Cloud Computer is a dedicated environment Manus sells for work that outgrows one machine. The blog names two uses: the server behind a multiplayer game, and a permanent home for an automation. Manus does not disclose Cloud Computer pricing in the launch post or the Help Center pages reviewed here. Purchases are real; the rate card is not on the public page.
Automations replace the old story where work started only on a schedule. Scheduled Tasks, in Manus’s wording, used to fire at a set time. Automations can also start when something happens in a connected service. The blog lists triggers including a new email, a change in ad performance, a calendar event, a Slack message, and a Notion update. The user describes what to watch and what to do next in one prompt, and Manus sets up the workflow. The Help Center example watches a product account on X and drafts a summary when a new post appears. The human still decides what goes out. The shift is from clock-driven runs to event-driven runs, not from human approval to full autopilot.
Video Editor also includes Alchemy mode, which Manus describes as AI acting as creative director from an idea, combining video and code generation. The blog claims a large quality jump in ideas, visual control, motion, and pacing, and says users can still fine-tune on the timeline afterward. Game Dev’s Max Ultra mode is named for higher-end visuals and feel. Those modes are marketing labels on the same Studio add-ons, not separate products with their own benchmarks.
Remote Control and Computer Use sit beside Studio on the desk. Remote Control streams the desktop to the phone while Manus executes voice commands, such as finding the latest deck in a launch folder and sending it. Computer Use is the underlying system: a visible, separate workspace on the authorized machine, limited to approved files, browsers, and apps. Away from the desk, the computer can keep working while the user stays in control of what it is allowed to touch.
Cue: email, phone, wallet, computer
Cue is a standalone app built on the same infrastructure as Manus. The blog tells users to “give it a cue” and let it work. Each agent receives its own email address, phone number, wallet, and computer. With that identity it can send messages, pay within a budget the user sets, finish tasks on its own machine, take calls, and leave a summary in Cue.
Teamwork is a group chat with a shared goal. Manus’s examples pass work between agents: one researches venues for a New York launch, another builds a shortlist, a third drafts a deck. The Help Center uses a marketing channel with three agents on the same chain. The user sets direction, checks in from the phone, and makes the final call. Cue also describes a restaurant scenario where scanning a QR code lets an agent order or hold a place in line.
Access is invite-only. The Help Center says to open the Agents (Cue!) tab in the Manus sidebar, scan a QR code to download Cue, and enter the code at sign-in. The launch post says early access is free with a code but does not publish pricing for Cue after early access ends. Manus does not disclose Cloud Computer rates in the same public materials, so the full cost of an agent that needs an always-on machine is still partly opaque.
Phone numbers are part of the identity story and part of the fine print gap. Manus promises each agent its own number and the ability to take forwarded calls. The launch post and Help Center do not include a list of countries where agent telephony works. Manus does not, in those documents, say which carriers, which number types, or which regions are supported. Until the company publishes that list, “select countries only” is the honest summary for readers outside Manus’s own marketing: the feature exists in the product description, and the geographic boundary is undefined in the primary sources.
What the launch leaves open
Manus 2.0 is broad on surfaces and narrow on proof. Cascade’s token, time, and cost improvements are real numbers on the page, tied to one configuration. Studio expands the file types and adds timeline editing and game templates. Automations move triggers from the clock to inbox, calendar, chat, and docs. Cue turns identity into infrastructure: inboxes, numbers, wallets, and computers that belong to agents rather than to a single login.
What remains undisclosed is what a buyer needs to model spend. Cloud Computer pricing is not on the launch post. Cue’s price after early access is not on the launch post. Agent phone coverage by country is not in the Help Center articles Manus published alongside the release. iOS for Cue is waiting on App Store review while web, desktop, and Android paths are live. The invite code MEETCUE is first-come with limited spots, which means the most visible feature of the day, agents that can pay and answer the phone, is still rationed at the door.
That rationing is the story beneath the architecture slide. Manus is selling a harness that loads tools on demand, a studio that keeps edits on a timeline, and a second app where agents carry their own contact details and payment limits. The thought experiment ended when the company asked for a code. The receipts, benchmarks, country lists, and price tags are still catching up.