Manus AI launches a general AI agent capable of handling complex real-world tasks, including creating video games

Manus AI has launched its new general AI agent capable of independently researching answers to complex prompts by leveraging multiple AI large language models (LLMs) in parallel. The AI is currently available by requesting an invitation.

Common chatbots, such as OpenAI ChatGPT, Microsoft CoPilot, and Anthropic Claude, are trained on a fixed set of data, so their knowledge is limited. Questions without answers in their training dataset cannot be answered, although some companies try to expand chatbots by allowing them to access the Internet for the latest information. Still, these chatbots cannot answer complex prompts that require problem solving.

Some AI companies have tried to tackle this by allowing the AI to think through problems step-by-step, analyze the data it finds online, and synthesize an answer. OpenAI Deep Research is one such AI agent that was released last month, and Manus AI is the newest.

Unlike OpenAI's offering, Manus's agent uses multiple AI LLMs, thus benefiting from the advantages each provides. Prompts are automatically split up into smaller tasks that are worked on in parallel. Users can follow the AI's thinking as it automatically starts working through problems step-by-step. The agent can produce not only text answers but also spreadsheets, interactive charts, webpages, and video games.

Although Manus AI's agent only scores 57.7% on Level 3 prompts in the GAIA AI benchmark, a test of real-world questions that even humans have difficulty answering, the AI agent is able to correctly answer easier Level 1 and 2 prompts more than 70% of the time. According to Manus AI, it performs better than other AI capable of researching answers today.