Zhipu AI is stepping up its role in China’s homegrown AI movement by offering its new agent system, AutoGLM Rumination, to the public free of charge via its official website and mobile platforms.
The move adds pressure to China’s increasingly competitive AI sector, where firms are racing to develop autonomous systems capable of both reasoning and execution—while avoiding dependency on foreign technology and cloud infrastructure.
As rivals introduce subscription models or gate access behind invitation codes, Zhipu’s open release stands out—not only for its accessibility but for the company’s broader roadmap, which blends technical ambition with geopolitical alignment.
AutoGLM: An Agent Powered by Task-Ready Language Models
AutoGLM Rumination builds on a sequence of language models developed by Zhipu: GLM-4, GLM-Z1, GLM-Z1-Rumination, and now the AutoGLM Rumination agent system. Unlike chatbot-style assistants, the agent is built to perform tasks such as generating research reports, searching the web, and planning travel.
Zhipu says the agent runs on its GLM-Z1-Air and GLM-4-Air-0414 models, which were designed with performance and efficiency in mind.
According to Zhipu AI, the GLM-Z1-Air model can match the performance of DeepSeek-R1 while running up to eight times faster and requiring only one-thirtieth of the computing power. A smaller variant, GLM-Z1-Flash, is also available for use on lower-end devices. AutoGLM is currently accessible for free through Zhipu Qingyan on the web, PC, and mobile, and the company plans to open-source its core components on April 14.
AutoGLM supports tool-based task execution, long-context reasoning, and browser automation—areas in which the company claims its models outperform GPT-4o and Claude 3.5 Sonnet in the Stanford-led AgentBench benchmarks.
Chinese AI Agents on the Rise: Zhipu vs. Manus
While Zhipu leans into openness and public benchmarks, another Chinese startup, Manus AI, has followed a very different approach. Manus just its paid plans which include a $39 per month Starter tier and a $199 Pro tier.
Access to Manus was so in demand during its invite phase that some codes reportedly sold for up to ¥50,000 (about $7,000 USD). Its agent is pitched as a fully autonomous system using LLM chaining, multisig control, and reinforcement learning to perform multi-step tasks with minimal user input.
Early testers praised its resume generation and layout skills but noted occasional hallucinations and missing source citations. These issues raise concerns about reliability, especially in mission-critical workflows.
Unlike Zhipu, which backs up its claims with public data and open availability, Manus keeps most performance metrics and model internals undisclosed. That strategic divergence may play into broader perceptions of transparency and accessibility in China’s agent race.
Strategic Backing and Roadmap Development
Zhipu AI was founded in 2019 as a spin-off from Tsinghua University and has steadily gained favor with both private and state-backed investors. In March 2025, the company secured 500 million yuan (around $69 million USD) in funding from the state-owned Huafa Group, adding to a previous 3 billion yuan round. Its valuation as of July 2024 was 20 billion yuan.
The roadmap for 2025 emphasizes specialization in agent-oriented foundation models—systems built to reason, operate tools, and carry out multi-step actions autonomously.
Zhipu also signaled its international ambitions with the formation of a coalition with representatives from 10 Belt & Road and ASEAN countries, with the goal of helping member states develop sovereign AI infrastructure. This aligns closely with China’s broader narrative of AI self-sufficiency and standard-setting abroad.
Outside the agent space, Zhipu has also demonstrated its capabilities in speech and multimodal AI. In October 2024, the company released GLM-4.0, an open-source large language model with voice synthesis and tone adaptation features. This further reinforced its focus on adaptable, locally deployable AI systems.
Still, experts caution that running these agents on limited hardware—despite the appeal of low compute costs—may introduce latency or task failures when chains of operations become complex. Benchmark data like AgentBench is helpful, but real-world edge deployments often reveal bottlenecks not captured in test conditions.
Yet the message from Zhipu is clear. By combining public access, open-sourcing plans, and global alliance-building, the company is betting on scale, accessibility, and alignment with China’s strategic tech objectives. AutoGLM isn’t just a chatbot; it’s a functional agent designed to operate independently—and it’s now in the hands of millions.
Policy Tensions and Global Scrutiny
Zhipu’s rise has not gone unnoticed by international regulators. Like other Chinese AI developers, it was blacklisted by the U.S. government over national security concerns. Despite this, it continues to receive robust domestic support and maintains strong partnerships with city governments and local tech firms.
The regulatory picture is even more turbulent for competitors like Manus. Following growing concerns over security, U.S. states such as Tennessee and Alabama banned the use of Manus on government networks, citing risks tied to censorship, propaganda, and cybersecurity threats. Federal discussions are underway about classifying autonomous agents like Manus and DeepSeek as high-risk platforms requiring oversight.
Federal agencies are weighing a broader AI platform policy that could include Zhipu’s models. These developments triggered an uptick in domestic chip hoarding in anticipation of further U.S. export controls.
