Manus My Computer: The Chinese AI Agent Challenging Claude Cowork

Manus My Computer: A Serious Competitor to Claude Cowork

Manus My Computer, developed by a Chinese company specializing in agent technology, positions itself as a direct rival to Claude Cowork, the AI agent from Anthropic. Launched in January, Claude Cowork quickly gained popularity by automating complex tasks for knowledge workers, notably through a partnership with Microsoft to integrate the Office suite. Claude Cowork has established itself as a benchmark for knowledge workers, focusing on the data aspect.

Less than two months after the launch of Claude Cowork, Manus introduced Manus My Computer, a clone focused on autonomy and speed. This new solution promises to compete with its San Francisco counterpart by offering impressive performance, although its cost may be a barrier.

How Manus Works and Its Features

To use Manus My Computer, the publisher developed a Desktop application, operating on the same principle as Claude Desktop. The AI agent takes control of your machine to execute complex tasks, using bash commands to analyze and modify files locally. It can also control your local applications through its computer use mode. Before performing an action, Manus requests user approval, unless default permission has been granted.

The application is available on macOS and Windows and can be controlled remotely via applications like Slack, Telegram, or Line. Manus goes even further than Claude by offering a turnkey API to create and launch tasks remotely. Use cases include sorting photographs, managing invoices, and even creating applications from scratch. A collaborator from the startup created a real-time translation and subtitling application in Swift in just 20 minutes.

Manus My Computer is marketed as an alternative to OpenClaw, although the latter is not explicitly named. The company recommends installing My Computer on a dedicated Mac mini, kept on 24/7, for optimal use.

The Manus Desktop agent is based on the Manus 1.6 agent family, which comes in three versions: 1.6 Light for simple tasks, 1.6 Classic for medium tasks, and 1.6 Max for complex cases. Unlike Anthropic, Manus only controls the scaffold of its agent. Although the models used are not disclosed, the subcontractors listed on the Trust Center, such as Anthropic and Google Cloud, suggest that Claude is the engine model, supplemented by the Gemini suite for other tasks depending on their complexity. Since the partners are based in the United States, personal data should not be sent to China.

A Credit-Based Pricing Model

Unlike Claude Cowork, which offers a fixed monthly subscription, Manus uses a credit system. Each action performed by the agent consumes a variable number of credits depending on its complexity. A simple file sorting task will cost a few dozen credits, while a complex workflow, such as creating an application or conducting in-depth multi-source research, can consume between 500 and 900 credits.

Three paid plans are offered: Standard at $20 per month for 4,000 monthly credits, Customizable at $40 for 8,000 credits, and Extended at $200 for 40,000 credits. All plans, including the free one, benefit from 300 daily refresh credits, access to all three versions of the Manus 1.6 agent, and 20 simultaneous tasks. A Team offer is also available, with a shared credit pool among members and administrative functions such as SSO and access controls. Annual billing offers a discount of about 17%. Although Manus appears aggressive on paper by offering more uses than Claude Cowork, the credit model introduces unpredictability that is absent with Anthropic.

Performance Test: A Machine Learning Challenge

To evaluate Manus's capabilities, we presented the agent with a daily machine learning challenge: to build a model capable of automatically classifying files in the Downloads folder. The agent was required to scan the folder, extract metadata from each file, such as extension, size, timestamp, and naming pattern, construct a usable dataset, train a classifier to predict the destination category of each file, and then apply the model to actually reorganize the folder.

We used the most advanced agent, Manus 1.6 Max, and provided the following prompt: "Analyze my Downloads folder and build a machine learning model capable of automatically classifying my files into relevant categories. Start by scanning the entire folder to extract metadata from each file, then create a structured dataset from this data. Perform feature engineering on the file names to deduce useful signals, such as the presence of dates, hashes, and keywords like invoice, contract, screenshot, IMG. Train a classification model using techniques like random forest or gradient boosting to predict the destination category of each file, evaluate its performance, then apply it to reorganize my folder by creating subfolders and moving the files. Produce a final report. And set up a sorting routine based on the model once a week."

Once the prompt was sent and permanent access permission to the Downloads folder was granted, Manus got to work. The agent began by creating a plan, then executed all necessary commands in a virtualized environment, an Ubuntu VM. It consulted the files, built a dataset, trained a classification model, applied the model, produced a report, and set up the requested weekly sorting routine. The first observation: Manus is much faster than Claude Cowork, completing the task in just about ten minutes.

The final result met expectations: the produced report was clean, structured, with model metrics and details of the files moved by category. The classifier correctly identified the vast majority of files, and the structure created in the Downloads folder was immediately usable. The destination categories included work documents, invoices and administrative files, photos and images, installers and archives, and temporary files to be deleted. In terms of credits, the task consumed 1,078 credits, which is more than a quarter of the monthly Standard plan for a single operation. With a few complex tasks per week, the credit budget can deplete very quickly.

An Excellent Agent, but a Pricing Model to Reconsider

At the end of our test, the conclusion is clear: Manus My Computer is, on paper, an excellent alternative to Claude Cowork. The Chinese agent gets the job done, and it does it quickly, very quickly indeed. The real downside is the credit consumption. With intensive use, the bill can quickly exceed that of a Claude Pro subscription at $20, or even rival the Max offer at $200, without the predictability that a fixed plan provides.

For companies that genuinely want to automate their processes, Manus still has significant advantages: its turnkey API, its architecture designed for deployment (dedicated Mac mini, remote control), and its near-total autonomy. We now hope that Manus will reconsider its pricing… Unless this pricing (increasingly common) simply reflects the real cost of agentic AI in 2026, which no one really wants to admit yet.