On March 6, 2025, a Chinese startup named Monica unveiled Manus, heralded as the “world’s first fully autonomous AI agent.” This isn’t just another chatbot designed to answer questions or generate text—it’s an AI built to think, plan, and act independently on complex real-world tasks. From crafting custom websites to analyzing stock portfolios, Manus promises a leap beyond the capabilities of familiar models like OpenAI’s ChatGPT and Deepseek. But what sets Manus apart, and how does it stack up against the competition? A key piece of the puzzle lies in its performance on the GAIA benchmark—a test that’s becoming the gold standard for evaluating AI’s practical problem-solving skills. Let’s dive into what Manus is, how it compares to heavyweights like OpenAI and DeepSeek, and why it’s generating such buzz.
What Is Manus?
Developed by a team led by tech entrepreneur Ji Yichao, Manus is a cloud-based AI agent designed to go beyond passive conversation. Unlike traditional models that wait for your prompts, Manus takes initiative. Ask it to plan a trip to Japan, and it might research flight prices, local weather, and cultural events, then deliver a detailed itinerary—all without you micromanaging the process. Need a website? It can code one from scratch. Curious about Tesla’s stock? It’ll analyze market trends and competitors autonomously.
This autonomy is Manus’s big selling point. It operates in real-time, browsing the web, using tools, and showing its step-by-step reasoning as it works. Early demos have showcased tasks like comparing insurance policies, screening resumes, and even finding apartments by factoring in crime rates and rental trends. Currently, it’s available only through an invite-only preview at manus.im, fueling both excitement and skepticism. Invite codes are reportedly fetching thousands of dollars on Chinese platforms like Xianyu, hinting at the hype surrounding this AI.
The GAIA Benchmark: A New Yardstick for AI
To understand Manus’s claimed superiority, we need to talk about the GAIA benchmark. Developed by researchers to test AI’s ability to handle real-world challenges, GAIA (General AI Assistance) isn’t about reciting facts or writing poetry—it’s about solving problems that require reasoning, research, and tool use. Think of questions like “What’s the cheapest way to ship a package from New York to Tokyo?” or “Which university offers the best AI program based on faculty publications and funding?” These tasks demand web searches, data synthesis, and decision-making—skills that mirror human problem-solving more closely than traditional benchmarks like MMLU (which tests factual recall).
GAIA scores AI models on accuracy, efficiency, and autonomy. A high score means the AI can independently find reliable answers without hand-holding. It’s a tough test, and until recently, even top-tier models struggled to excel here. OpenAI’s ChatGPT, for instance, shines in conversation but often falters when tasks require real-time web access or multi-step planning without explicit instructions. This is where Manus claims to shine.
Manus vs. OpenAI, DeepSeek
Monica’s bold claim is that Manus outperforms OpenAI’s DeepResearch—an advanced research-focused model—on GAIA. DeepResearch, built on GPT architecture, excels at digging into complex queries but still leans on user guidance for direction. Manus, by contrast, reportedly takes the reins entirely, planning its approach and executing it with minimal input. Testers, including Hugging Face’s head of product, have praised its fluidity and initiative, calling it a “game-changer” in AI autonomy.
How does it compare to OpenAI’s broader offerings, like GPT-4 or the upcoming GPT-5? OpenAI’s models are powerhouses in natural language processing, generating human-like text, and excelling in creative tasks. However, their real-world task execution is limited by a lack of native autonomy and real-time web integration (though plugins and updates have narrowed this gap). Manus’s edge lies in its ability to act as a self-directed assistant, not just a responder. If you ask GPT-4o to “find me an apartment,” it might list criteria or suggest websites; Manus, allegedly, delivers a shortlist with pros and cons, sourced and analyzed on its own.
Then there’s DeepSeek, another Chinese contender that debuted earlier in 2025. DeepSeek, impressed with its cost-efficient, high-performing language models, rivals Western AIs in technical tasks like coding and math. But DeepSeek remains a traditional LLM—great at processing prompts, less so at independent action. Manus’s developers argue it leapfrogs DeepSeek by prioritizing agency over raw computational power, though direct head-to-head GAIA scores for DeepSeek are less publicized
.
| Aspect | Manus AI | OpenAI DeepResearch | DeepSeek R1 |
| Core Strength | End-to-end task execution | Multi-step research synthesis | Mathematical/logical reasoning |
| Autonomy | Fully autonomous cloud operation | Requires user-guided prompts | Specialized problem-solving |
| Output | Functional files, screenshots, actions | Structured reports with citations | Analytical solutions with steps |
| Benchmark Performance | Outperforms DeepResearch on GAIA | Strong in research-focused tasks | Not specifically tested on GAIA |
Does Manus Outperform Chat GPT and Deepseek?
The claim that Manus beats DeepResearch on GAIA is tantalizing, but details are thin. Monica hasn’t released full benchmark data, citing the preview phase and server limitations. Independent verification is tricky with access restricted to invitees. Some skeptics wonder if the hype is inflated—perhaps a marketing ploy to ride China’s AI wave following DeepSeek’s success. Others see a genuine breakthrough, pointing to China’s growing AI talent pool and investment (the country’s AI sector saw $20 billion in funding in 2024 alone).
Anecdotal evidence supports the hype. Users report that Manus tackles tasks—like building a website in under an hour or analyzing a stock with real-time data—that leave ChatGPT or DeepSeek in the dust. Its real-time workflow transparency also builds trust, showing how it reaches conclusions. Still, without widespread testing, it’s hard to say Manus is the undisputed champ. OpenAI’s scale, DeepSeek’s efficiency.
The Bigger Picture
Manus arrives at a pivotal moment. China’s AI ambitions are surging, challenging U.S. dominance. If Manus delivers as promised, it could be a “Sputnik moment”—a wake-up call for Western AI labs to accelerate autonomous agent development. For users, it hints at a future where AI isn’t just a tool but a partner, handling life’s complexities with minimal oversight.
Yet, questions linger. How scalable is Manus? Can it maintain accuracy across diverse tasks? And what about ethics—does its autonomy raise risks of misuse? For now, it’s a tantalizing glimpse of what’s possible, backed by strong GAIA performance claims but shrouded in limited access.
Conclusion
Manus isn’t just another AI—it’s a bold step toward autonomous intelligence. Its reported edge on the GAIA benchmark over OpenAI’s DeepResearch and its leap beyond DeepSeek and others suggest a paradigm shift. Whether it truly outperforms remains to be seen as more users test it. One thing’s clear: Manus has sparked a conversation about what AI can be—not just a responder but a doer. As the preview expands, we’ll learn if it’s the future or just a flashy promise. What do you think—hype or history in the making?