After token revenue soared 760%, the market is re-evaluating Cloud Intelligence (09678)

Yun Zhisheng (09678) is entering a moment worth re-evaluating.

Let's look at a set of numbers first.

In the first half of 2026, Yunzhisheng achieved revenue of 562 million yuan, up 38.7% year on year; Big Model Token business revenue was close to 30 million yuan, up 760% year on year; of these, revenue in the second quarter exceeded 25 million yuan in a single quarter, an increase of more than 500% month on month.

What is more noteworthy is that this round of rapid token revenue growth is mainly driven by US dollar revenue.

This changed the meaning of the 30 million yuan figure.

If the growth mainly comes from traditional domestic projects, then it is more reflected in order expansion; while the increase in dollar token revenue corresponds to a different set of business models - users directly call the model API and pay according to token usage, and the model capability itself begins to become a commodity.

At the same time, the gross margin of the Token business has exceeded 60%, and the company's repurchase revenue accounts for more than 60%. On the other hand, Yunzhisheng is still increasing R&D. In the first half of the year, R&D investment reached 284 million yuan, an increase of 69% over the same period, but losses continued to narrow during the same period.

When a few curves are put together, a change that wasn't obvious before began to surface:

Yun Zhisheng is gradually revealing the revenue characteristics of a large model company from an enterprise that is commonly measured by the market using the logic of an “AI project company”.

This is probably where this interim report is really worth watching.

Over the past few years, the big model industry has been answering one question:

Whose model is stronger?

By 2026, the capital market will begin to ask the next level of questions:

Whose models are actually used by people? Who can get into the real workflow? Who can turn model calls into continuously growing revenue?

Looking at Yun Zhisheng from this perspective, revenue growth of 38.7% is not the most interesting figure in this financial report.

The real thing worth watching is Token.

Because it is providing the market with a yardstick to re-evaluate this company.

The token curve, which is driven by dollar revenue, has become steeper due to the new standard

Over the past six months, YunZhisheng's most intuitive change came from Token.

In the first half of the year, Big Model Token's business revenue was close to 30 million yuan, up 760% year on year; of these, the second quarter's revenue in a single quarter exceeded 25 million yuan, an increase of more than 500% month on month. In other words, the vast majority of token revenue in the first half of the year was concentrated in the most recent quarter, and the growth curve is clearly steeper.

And this round of growth is mainly driven by dollar revenues.

For a big model company, this is important.

Because tokens have completely different business logic from traditional AI projects.

Traditional software and AI projects rely more on procurement, customization, and delivery, and each contract corresponds to a revenue; token revenue is only generated when the model is actually invoked.

You only earn money when someone calls; the more calls, the higher the revenue.

Therefore, the reflection behind token revenue is not only that the model has been “sold”, but that the model is being continuously used.

Especially after entering the Agent era, this logic will be further amplified.

In the past, humans and AI mainly answered questions one by one, and one question corresponded to a small number of model calls. But as agents begin to perform complex tasks, a workflow may be broken down into dozens or even hundreds of steps, each of which may call on models, tools, or external systems.

The unit that consumes models is shifting from “one conversation” to “accomplishing one thing.”

Yunzhisheng U2 can independently dismantle and advance complex workflows of more than 100 steps, essentially adapting to this change.

Therefore, what is really worth watching is not the 30 million yuan itself, but the nature of the income behind this curve.

Currently, the gross margin of the Token business has exceeded 60%. This means that Yunzhisheng is not simply relying on low prices in exchange, but is forming a new revenue curve with high growth, high gross profit, and recurring characteristics.

Project revenue is more like “selling once.”

Token revenue means “continuous use.”

The business logic of an AI company only really began to change when the model began to make money based on usage.

A change from a “project” to an “asset”

But tokens are just results.

What really supports this curve is another change that YunZhisheng is taking place.

In the past, the typical perception of Cloud Intelligence from the outside world was that it was an AI solution company deeply involved in the fields of medical care, health insurance, and the Internet of Things.

The advantages of this model are obvious: deep understanding of the industry, high customer stickiness, and strong entry barriers.

However, the traditional project system also has a natural problem:

More work on a project often means more invested people.

And the big model is changing that.

At present, Yunzhisheng's intelligent enterprise services account for more than 60% of repurchase revenue, and the scale of repurchases has increased by 40% over the same period last year. At the same time, the company is disassembling and standardizing the business capabilities accumulated in various projects in the past, and then sedimentation it into intelligent modules that can be used repeatedly.

It used to be:

One customer, one demand, one set of projects.

Now it gradually becomes:

Model Base+Standard Capability Module+Agent Orchestration.

The biggest difference is that after a project is over, competencies don't end with the project, but can continue to accumulate and become the foundation for the next delivery.

As a result, the logic of growth also began to change.

It used to be that the more projects, the more people.

In the future, it could become:

The more projects, the richer the reusable models and agent capabilities, and the faster the next delivery.

This is probably the second key to understanding the changing value of YunZhisheng.

It is gradually shifting from earning money from engineering capabilities to model capabilities and reuse of business assets to make money.

Healthcare is not just an industry market

Where do these reusable abilities come from?

Healthcare is an important answer.

As of the first half of 2026, Yunzhisheng has served more than 470 medical institutions, of which more than 80% are tertiary hospitals. In the medical record quality control scenario, AI can reduce the single review time to less than 10 seconds, improving manual quality control efficiency by 80%; in the Jiangsu Health Insurance Intelligent Management System, manpower investment has been reduced by about 65%.

But if you only understand these as “industry cases,” you actually underestimate the significance of healthcare for a large model company.

What makes healthcare really special is:

It's complicated enough, and serious enough.

Writing a copy allows for certain deviations.

However, medical records, medical insurance reviews, and auxiliary diagnosis and treatment are difficult to accept as “almost correct.”

How can an intelligent body dismantle a task? Which steps can be automated? Which nodes must be confirmed manually? What mistakes must never happen?

These problems are difficult to solve with Benchmark alone.

They have to go into the real world.

Therefore, for YunZhisheng, healthcare is becoming a “training ground” for models and intelligence from the “business scenario” of the past.

The model enters the real scene to solve the problem; the real scene in turn helps the model to develop more complex execution capabilities.

This is also where “strong base model x deep application” is really worth watching.

The application is not the end point of the model, but it can also be the starting point for the model to continue to evolve.

Model competition is also beginning to change the scale

All commercial changes eventually have to go back to the model itself.

Even the best business model is difficult to establish if the underlying model isn't strong enough.

In June of this year, Yunzhisheng released a large model of the U2 native smart body. Its core direction is not simply to pursue a larger parameter scale, but rather emphasizes complex task execution and “high intelligence density.”

The logic behind it is straightforward:

The future of AI competition depends not only on whose model is bigger, but also on —

Exactly how much effective intelligence can be generated per unit of computing power.

Especially after entering the age of agents, this issue will become even more important.

A task may call the model dozens or even hundreds of times in a row, and the cost of inference will be amplified each time.

Therefore, commercializing big models is ultimately a multiplication problem:

Model ability × inference efficiency × scene value.

It's hard to set up less than one.

What is interesting about Yun Zhisheng's interim report is that these variables began to rotate simultaneously for the first time:

Improved model capability, bringing agent execution capability;

The agent enters the actual workflow and generates more token calls;

Token calls have expanded, and they have begun to be converted into high gross income again;

Real industry projects continue to accumulate, further forming models and business assets that can be reused.

A commercial flywheel is beginning to appear.

This has also made the loss narrower, which is worth re-understanding.

R&D is still growing rapidly, but commercialization has begun to accelerate at the same time.

If this trend continues, for a large model company, scale will not only be a cost, but may also be a lever for revenue and profit.

More important than the number of tokens is the value of tokens

This also brings up a more interesting question:

Should Big Model Companies Really Be Pursuing More Tokens?

Not necessarily.

If a token only completes a simple chat, it has limited commercial value.

However, if a string of tokens completes a medical record quality control, a medical insurance review, or even promotes the execution of a complex business process, its value is obviously completely different.

Therefore, what the big model industry really needs to compete next is probably not just the number of tokens.

but rather:

How many problems did each token solve.

This also corresponds to the two concepts that YunZhisheng has always emphasized:

Intelligent density, and token value.

The former determines how much intelligence a unit of computing power can generate, while the latter determines how much real value these intelligences can ultimately create.

The combination of the two is the real commercial efficiency of an AI company.

In the past, the market looked at YunZhisheng and saw medical, Internet of Things, chips, voice, and AI projects one by one.

In the age of big models, these originally seemingly scattered abilities are coming back together:

The model provides a smart dock;

Real scenarios provide high-value tasks;

Agent connects to business workflows;

Tokens are starting to turn intelligence into revenue.

As a result, the capabilities that a company has accumulated over the past ten years began to gradually point to the same business logic.

China's big model industry has been telling the story of technology for three years.

By 2026, the market is becoming increasingly concerned about another thing:

Can technology actually be implemented.

Whose models do people actually continue to use?

Who can get into the real workflow?

Who can turn intelligence into high-margin, repurchasable, and continuously growing revenue?

Seen from this perspective, the fact that Yun Zhisheng's interim report is really worth paying attention to is not that a single number suddenly becomes beautiful.

Instead, a few things that were difficult to establish at the same time in the past are happening together:

The model continues to be invested, but commercialization has begun to accelerate; the industry's business continues to grow, and dollar token revenue has begun to climb steeply;
projects are still being delivered, but they have begun to settle into intelligent assets that can be reused over and over again.

In the past, the market understood YunZhisheng more as an industry solution company with AI technology capabilities.

Now, this set of scales may need to be adjusted.

Because when the model begins to directly generate revenue, when one-time deliveries begin to be converted into continuous use, and when industry know-how begins to become reusable assets, the value logic of a company changes accordingly.

What YunZhisheng actually welcomed was probably not a “shift in gear” in business.

It was a time for the market to re-evaluate.