①Zhang Peng pointed out that the keyword for 2026 is called 'Token Volume.' Applications represented by OpenClaw have triggered a frenzy of Token consumption. In response to the supply shortage since February, the company will continue to increase investment in the optimization of domestic chip hardware-software integration to push inference performance to its limit; ②Token exports are not about low-price competition but rather high-quality and premium-priced output based on models like GLM-5.
$KNOWLEDGE ATLAS (02513.HK)$ Released its first annual report card following the initial public offering.
The financial report shows that in 2025, the company achieved a total revenue of 724 million RMB, representing a year-on-year increase of 131.9%, making it the largest large-model company in terms of income domestically. More notably, its MaaS API platform ARR reached 1.7 billion RMB (approximately 250 million USD), increasing 60-fold year-on-year, with gross margin improving nearly fivefold to 18.9%.
Today, after the release of the earnings report, Zhipu's share price surged by over 30% at one point. As of the time of writing, it has risen by 31.87%.
In the view of Zhipu CEO Zhang Peng, the underlying logic of this growth is precisely the essence of commercial value in the AGI era — the upper bound of intelligence multiplied by Token consumption scale. During the earnings call, he told reporters from The STAR Market Daily that Zhipu’s goal is to become the infrastructure that enhances society-wide TAC, ensuring every unit of Token can be transformed into deliverable economic growth.
Token Economics: From 'Invocation' to 'Productivity'
Zhang Peng repeatedly emphasized a key term in his speech — 'Token Volume.' He stated that the keyword for 2026 is 'Token Volume.' Applications represented by OpenClaw have ignited a surge in Token consumption. Facing a market with supply shortages since February, the company will continue to increase investment in optimizing the integration of domestic chips' hardware and software to push inference performance to its limits. This effort is not for short-term profits but to support the continually rising curve of high-quality Token consumption.
“The medium through which intelligence benefits the public is Token. Every individual and organization can invoke intelligence via Token to generate value.” Zhang Peng said.
Under this philosophy, Zhipu defined the value measurement standard for AI productivity — Token Architectural Capability (TAC), i.e., 'Intelligence Invocation Volume × Intelligence Quality × Economic Conversion Efficiency.' In the future, the benchmark for measuring individual or organizational value will no longer be the amount of information possessed but the ability, as a Token Architect, to construct complex Agent systems and drive autonomous model operations within a given budget.
As the upper bound of intelligence breaks through and Token consumption accelerates, between 2025 and 2026, Zhipu completed multiple iterations of its models, from GLM-4.5 to GLM-5-Turbo.
The significant enhancement in model capabilities has directly driven exponential growth in Token consumption. Zhang Peng stated that as models become more powerful, user scenarios grow deeper and more complex, leading to an increase in Token usage. Anthropic's annual recurring revenue (ARR) grew tenfold over the past year, while Zhilu API platform surged sixtyfold, both reflecting that growth is no longer linear.
Notably, the earnings report mentioned that despite an 83% price increase for APIs, the market remains undersupplied. However, a question arises: Is this pricing power due to GLM’s genuinely leading model capabilities or a temporary seller's market caused by current computational power shortages? If the computational gap eases, can such pricing levels be sustained?
In response, Zhang Peng told reporters from The Science and Technology Board Daily that he first acknowledged the recent constraints and bottlenecks in computational supply. However, in the broader market environment, many vendors provide API services based on computational power. Clients accepting Zhilu's price hike and continuing to choose its services demonstrates recognition of Zhilu’s model capabilities.
He believes that the essence of long-term pricing is determined by value. Resources that effectively replace human labor, improve conversion efficiency, and enhance intelligence levels are scarce and valuable. The company focuses more on the value created per Token and the value delivered to clients. Only when clients recognize this value will they be willing to incur higher costs. “Thus, I believe our pricing power is determined by our technological strength and the leading position brought by long-term trends.”
Moreover, he noted that the sixtyfold increase in API usage over the past year marks the beginning of a structural, long-term trend. Today, AI capabilities have evolved from being functional and entertaining to solving increasingly complex and critical problems, transforming Token API calls and consumption into tangible economic value.
“Additionally, the emergence of new application forms like OpenClaw and expectations for device-level native intelligence indicate that future API and Token consumption will grow exponentially. We believe the growth in API ARR represents a definitive, long-term trend.”
According to reports, Zhilu’s MaaS platform, BigModel.cn, has connected over four million enterprises and developers, covering 218 countries and regions worldwide. Nine out of China’s top ten internet companies now deeply integrate GLM daily.
Meanwhile, Zhilu launched the 'GLM Coding Plan' programming package domestically in 2025, with the number of paying developers surpassing 242,000, and Token usage increasing fifteenfold within six months. In March 2026, following the Coding Plan, the 'Claw Plan' was introduced, gaining over 400,000 subscribers within 20 days of launch, validating the commercial potential of agent-based long-chain tasks.
From 'Made in China' to 'Intelligent China' Export
Zhilu’s Token economy is not limited to the domestic market.
It was introduced that currently, the GLM model is also deployed on global cloud service providers such as Google Vertex AI, AWS Bedrock, Fireworks, and Cerebras, and has become the top-ranked paid model on OpenRouter.
Zhang Peng stated that relying on China's full industrial chain advantages in energy, chip adaptation, and IDC operations, Zhipu is achieving a leap from 'Made in China' to 'China Intelligence.' 'The overseas expansion of Tokens is not about low-price competition but rather the high-quality and premium-priced output based on top-tier intelligence like GLM-5. What we aim to supply globally are production factors that represent the upper limit of cognitive intelligence with ultimate cost-performance.'
Looking towards 2026, Zhipu has proposed two major strategic directions:
The first is targeting the TAC era, where everyone becomes a 'Token Architect.' Zhipu's goal is to become the infrastructure that enhances society's overall TAC capabilities, ensuring every unit of Token can be transformed into deliverable economic growth.
The second is betting on LLM-OS (Large Language Model Operating System), transitioning from a 'dialogue interface' to an intelligent scheduler. The future computing platform will no longer be a stack of Apps but a synergy between API stores and Agent matrices. Zhipu is committed to making GLM the core engine of this autonomous system, enabling a leap from cloud-based APIs to device-level native intelligence.
"The journey toward AGI is long, and we will continue to work diligently, using our ultimate foundational capabilities to explore unknown boundaries." Zhang Peng stated, "Zhipu is not a traditional software company; it will continue along the commercial path of being 'China’s Anthropic,' acting as a native intelligence laboratory with AGI as its belief."
Looking to pick stocks or analyze them? Want to know the opportunities and risks in your portfolio? For all your investment-related questions,just ask Futubull AI!
Editor/Doris