eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
A development team based in China recently unveiled an autonomous AI agent known as Manus. While many of the details remain scarce, their solution has reportedly outperformed OpenAI – one of their primary competitors – in early benchmarks.
What is an AI agent?
Designed to make informed decisions, perform basic tasks, and learn based on their prior experiences and interactions, AI agents represent the next generation of virtual assistants.
Manus is an autonomous AI agent. Whereas many current AI models require human interaction via text-based chat or voice command, Manus functions without the need for step-by-step instructions. Capable of operating independently, some of Manus’ initial capabilities include sourcing B2B suppliers, mapping potential customers, creating educational materials, and travel planning.
While AI agents aren’t exactly new, the rise of large language models (LLMs) has boosted their popularity. When used in tandem, AI agents and LLMs make it easier to interact with AI and achieve specific objectives.
Analyzing the early benchmarks
Known as Manus AI, the Chinese-based development team behind the new AI agent posted an introductory video on YouTube to announce their latest innovation. Not only did it cover some typical use cases, such as resume screening, real estate property research, and stock analysis, but it also highlighted Manus’ results in early benchmark tests.
They used GAIA, a common benchmarking system for AI assistants and other generative AI tools, to test Manus’ ability in solving real-word problems. When compared to previous benchmarks that were once considered “state-of-the-art” (SOTA), Manus scored higher on all three difficulty levels.
But the tests didn’t stop there. The Manus AI team also measured Manus’ performance directly against OpenAI, and Manus bested those numbers, too.
- Level 1: Manus (86.5%) / OpenAI (74.3%) / Previous SOTA (67.9%).
- Level 2: Manus (70.1%) / OpenAI (69.1%) / Previous SOTA (67.4%).
- Level 3: Manus (57.7%) / OpenAI (47.6%) / Previous SOTA (42.3%).
Their initial benchmarks are promising, but Manus is currently only available through an invitation-only preview. An exact release date has not been announced, and it’s unclear when the new AI agent will be available to the general public.
Ramping up the AI models competition
Competition in the current AI landscape is increasing every day. Between Manus and several other types of AI models that have been released in 2025, the year is already shaping up to be a pivotal time in the development of next-gen AI.