AI Agent Platform Manus is All the Hype
China’s Manus AI is quickly gaining traction among tech enthusiasts, with some touting it as a serious contender to OpenAI’s Deep Research and Operator agents.
However, its true capabilities remain uncertain, as access is tightly restricted to invite-only users—some reportedly paying exorbitant sums for entry.
Even among those who have tested the platform, opinions are divided on whether Manus is truly groundbreaking or merely overhyped.
What is Manus?
Manus, an AI agent platform developed by Beijing-based startup Monica AI, launched earlier this month, positioning itself as a versatile “general AI agent.”
Capable of booking travel, analysing stocks, generating educational content, and comparing insurance policies, Manus aims to handle complex tasks autonomously—eliminating the need for API configurations or intricate setups.
It interprets prompts, navigates the web, analyses on-screen data, and executes tasks seamlessly.
Unlike OpenAI’s advanced agents, which cost $200 per month, Manus is currently free but will introduce a subscription fee upon full rollout.
Limited server capacity has fuelled scarcity, leading to a secondary market for invite codes on China’s Xianyu marketplace, where beta access reportedly reached 10 million yuan ($1.3M) before being removed—potentially as part of a hype-driven FOMO strategy.
Named after the Latin word for “hand,” Manus markets itself as a “universal hand” for users.
The company claims it outperforms OpenAI and other agentic models across all difficulty levels on the GAIA benchmark, a key test for autonomous AI capabilities.
Manus Not DeepSeek’s Rival
User feedback on Manus has been mixed.
Andrew Wilkinson, co-founder of technology holding company Tiny, was among those who tested the model and praised its performance in initial trials.
AI researcher Victor Mustar echoed Wilkinson’s praise.
Meanwhile Deedy Das, an investor at Menlo Ventures, also expressed enthusiasm for the model.
However, scepticism has been growing, particularly after scammers launched a cryptocurrency token under the same name, temporarily leading to the suspension of Manus’s X account.
App developer AlvinWeb3 argued that Manus’s only misstep was becoming too popular too quickly.
Still, some users remain unconvinced, suggesting that the platform’s rapid rise is more of a marketing spectacle than a true breakthrough.
One user posted:
"Manus seems to have enlisted many Chinese AI influencers to praise it… Chinese netizens realized that this was a huge marketing scam, and Manus's reputation in China has been ruined."
Not all users were impressed with Manus.
Some found it sluggish, prone to looping, and inconsistent in executing complex tasks.
Pseudonymous user Teortaxes shared his disappointment on social media, highlighting the platform’s shortcomings.
Others were dismayed to learn that Manus relies on Claude rather than operating on a proprietary foundational model.
Monica AI’s Jichao Yi admitted:
“We use Claude and different Qwen-finetune.”
While the company integrates fine-tuned models and proprietary techniques, its core functionality appears to depend largely on existing AI frameworks, raising doubts about its originality and true agentic capabilities.
Open Source is the Solution
Given Monica AI's reliance on third-party models, some speculated that an open-source alternative would soon emerge—and they were right.
OpenManus, an open-source project developed by contributors from the MetaGPT community, offers a free alternative for those unwilling to pay $200 or wait for a Manus invitation.
This platform allows users to run AI agents locally with their preferred models.
The model’s card on AI Share Net reads:
"Compared to Manus, which requires an invitation code, OpenManus does not require any entry barrier, and users only need to clone the code and configure the LLM API to get started quickly."
While AI agents hold great promise, they have yet to prove their ability to handle truly complex tasks.
OpenAI is reportedly developing specialised agents that meet such demands—for a steep $20,000 per month.
For now, users face three choices: pay the premium, wait for access, or build their own.