Zhipu AI, a Chinese artificial intelligence startup, unveiled a new AI agent called AutoLM Rumination on Monday, Reuters reported.
The tool can perform deep research tasks, generate detailed reports, and help users plan various activities and tasks—all for free.
The AI agent is powered by the company’s proprietary models, which are claimed to be as good as DeepSeek R1.
The report also noted that the company claims its model operates eight times faster while using only one-thirtieth of the resources.
Recently, the startup secured $69 million in Series D funding, just weeks following a $137 million funding round. Zhipu is currently valued at $2.74 billion, and it has also been previously backed by giant companies Tencent and Alibaba. The startup, on multiple occasions, has been backed by Chinese state-owned entities.
In January, the US commerce department placed Zhipu and its subsidiaries on the export control entity list, preventing them from obtaining US-made components.
The announcement comes just weeks after Manus AI, an AI agent from a Chinese company, surprised many in the AI ecosystem. This “general-purpose AI agent” outperformed OpenAI’s deep research tool on the GAIA benchmark, which tests AI agents on solving real-world problems.
Manus AI is now available for a monthly fee of $39, along with an upgraded plan that costs $199 per month.
The AI ecosystem in China is currently experiencing unprecedented growth. Giants like Alibaba and Tencent are unveiling new AI models at an extraordinary pace. Recently, Alibaba introduced QwQ 32B, an AI model with 32 billion parameters. However, it is reported to achieve performance comparable to DeepSeek-R1, which has 671 billion parameters (with 37 billion activated parameters).
The model also surpasses OpenAI’s o1-mini in several benchmarks related to code, mathematical reasoning, and general problem-solving tasks.
While DeepSeek caused quite a storm in the markets a few months ago, the company doesn’t plan to stop. The company updated the DeepSeek-V3 model, and the DeepSeek V3-0324 now ranks highest in benchmarks among all non-reasoning models.
Reuters also recently reported that DeepSeek plans to release R2 “as early as possible”. The company initially intended to launch it in early May but is now contemplating an earlier timeline.