UBS evaluates Zhipu—“China's Anthropic”
A Tsinghua-affiliated AI company is taking a path highly similar to that of the world’s leading AI labs.
On April 20, UBS analyst Wei Xiong and others released a nearly 40-page research report, covering Beijing AI company Zhipu for the first time, giving it a “Buy” rating with a target price of 1160 HKD.
The report's core assertion is direct: Zhipu’s model development and commercialization path are highly similar to Anthropic—the world-leading AI company. Therefore, the analysts position Zhipu as the "China's Anthropic."
Why is it China's Anthropic?
Anthropic is one of Silicon Valley’s most prominent AI giants. It doesn’t pursue flashy video generation, but focuses on one thing: making AI write code.
UBS believes that Zhipu’s technological route and monetization method are nearly identical to Anthropic’s.
First, the strategic focus is highly aligned. Both companies chose programming capabilities as their breakthrough, because programming tasks are easy to verify, value is quantifiable, and it’s the shortest path for AI to transition from “assistance” to “execution.”
Anthropic bets on coding ability at the model level, emphasizing long-horizon tasks—that is, whether AI can consistently complete hours-long complex engineering tasks without intervention.
Zhipu has taken an almost identical path. As early as September 2022, Zhipu launched the AI coding assistant product CodeGeeX, becoming one of the first companies in China to focus on AI programming.
According to METR (Model Evaluation & Threat Research), Anthropic’s Claude Opus 4.6 can complete tasks equivalent to 12 hours of human work (50% success rate); Zhipu GLM-5.1 achieves about 8 hours, ranking first among open-source models globally.
Second, the performance gap is continuously narrowing.
According to Artificial Analysis data, as of April 17, 2026, Zhipu’s latest flagship model GLM-5.1 ranks seventh globally in comprehensive AI intelligence index with a score of 51.4, just 2 points less than Anthropic’s Claude Opus 4.6 (53 points).
In SWE-bench Pro (an authoritative test simulating real coding engineering tasks), GLM-5.1 scored 58.4, ranking second worldwide, only behind Anthropic’s Claude Mythos Preview (77.8), and beating GPT-5.4 (57.7).
Third, monetization pace is similar, but Zhipu’s ARR grows faster and offers better value.
In December 2025, Zhipu’s open platform annual recurring revenue (ARR) was $39 million. By March 2026, this had grown to $250 million—a 6.4-fold increase in four months.
In comparison, Anthropic achieved a similar 6.4-fold increase in its early days, but took about nine months. Recently, Anthropic’s value grew from $9 billion to $30 billion in just four months, but that’s a 3.3-fold increase.
Both companies monetize quickly, but Zhipu has the edge. The report states, “Compared to global leaders, Zhipu’s ARR growth is steeper, despite a smaller base.”
Meanwhile, Zhipu offers better value for money: GLM-5.1 is priced at about $2 per million tokens, while Claude Opus 4.6 with similar performance is about $9—Zhipu’s price is about 22% of Anthropic’s. Zhipu’s model offers higher value, meaning there is room for price increase over time.
In summary, at the model layer, both focus on programming rather than multimodality; at the engineering layer, both self-develop coding agents; and in commercialization, both price by model intelligence—the stronger the capability, the higher the price, and after price increases, demand rises rather than falls.
Programming: The Core Battleground
To understand Zhipu, one must first understand why it regards coding ability as its most important strategic pivot.
The report’s logic: Programming ability is the key threshold for AI entering real enterprise scenarios. If it can write code, it means AI can execute complex multi-step tasks, replace real engineering workflows, and greatly increase business value.
Zhipu has walked this path for nearly four years.
From CodeGeeX 1 in 2022 (a 1.3 billion parameter multilingual code model) to the GLM-5.1 released in April 2026, Zhipu has completed the leap from “code completion tool” to “independently running for 8 hours to complete complex engineering tasks.”
Specifically, GLM-5.1 achieved about 8 hours in METR’s long-horizon task test (AI completes tasks equivalent to 8 hours of human work at 50% success rate), ranking first among open-source models globally. In comparison, Anthropic’s Claude Opus 4.6 achieves about 12 hours, ranking first worldwide.
In Code Arena’s agent programming task rankings, GLM-5.1 ranks third globally, only behind Claude Opus 4.6 (Thinking version) and Claude Opus 4.6.
External benchmarking and industry research indicate that the GLM series models have become one of the “top choices” for corporate clients in programming-related tasks, with high visibility.
Pricing Power: Prices rise 30%, usage triples
The most direct test of commercialization is pricing power.
In February 2026, Zhipu directly raised coding plan prices by 30%, while tightening usage limits and model choices.
The result was unexpected: According to OpenRouter data, in March 2026, Zhipu’s total token usage soared about three-fold month on month.
Even more interesting: GLM-5.0 was the highest-priced model in Zhipu’s products at the time, yet its usage ratio in March 2026 was the highest among all models.
The report interprets: “This shows users’ willingness to pay for high-performance models and resilient demand; higher prices did not suppress usage.”
Looking at pricing evolution, Zhipu’s price increases follow two lines: one, synchronized with model upgrades—GLM-5.0 input price increased by 100% vs GLM-4.7, output price up 125%; GLM-5-Turbo and GLM-5.1 each increased about 20% over their predecessors. The other is direct price adjustment, like the 30% coding plan increase in February 2026.
Compared to Anthropic, Zhipu still has a significant value advantage. The report shows: GLM-5.1 and Claude Opus 4.6 have comparable intelligence scores (97%), but GLM-5.1’s pricing is only 22% of Anthropic’s. This means, as Zhipu’s model capabilities improve, its pricing has room for further increase.
R&D Strength: Tsinghua DNA
Zhipu’s R&D culture comes from Tsinghua University.
All five founders are Tsinghua alumni, Tsinghua University Asset Management Company holds 3.53% of Zhipu’s shares. Zhipu maintains deep cooperation with Tsinghua’s Knowledge Engineering Research Group.
The R&D team has over 800 people, R&D staff make up 74%, more than 20 papers published on arXiv (core preprint platform in AI), all above domestic peer labs. As of June 30, 2025, Zhipu's research and academic advisory team has published about 500 high-impact papers.
In technical innovation, the report highlights three key points: dynamic sparse attention mechanisms (similar to DeepSeek, reduces training/inference cost while retaining long-text capability), "Slime" asynchronous reinforcement learning framework (improves post-training efficiency), and native agent integration design (ARC framework, integrating agents, reasoning, and programming ability).
Two-pronged approach: On-premise deployment & open platform
Zhipu’s current revenue structure is “one heavy, one light”—on-premise deployment accounted for 74% of total revenue in 2025, open platform API 26%.
On-premise deployment: The stable foundation
On-premise deployment mainly addresses governments and large enterprises, with clients including nine of China’s top ten internet companies, as well as Hangzhou Urban Construction Investment Group, Zhuhai Huafa Group, Chengdu High-Tech Zone, and other government-backed entities.
In 2025, Zhipu won 57 large model projects in the public services sector, with a total contract value of 254 million RMB, ranking fifth among domestic large model vendors (the top four: iFlytek, Baidu, Volcano Engine, Aliyun).
On-premise deployment gross margin remains between 50% and 70%, 49% in 2025, considered healthy. However, the report also notes that days sales outstanding (DSO) rose from 107 to 153 days, accounts receivable grew from 100 million to 339 million RMB—this warrants continued attention.
Open platform: The growth engine of the future
The report is more optimistic about the open platform and API business.
In September 2025, Zhipu launched the coding plan; in March 2026, it rolled out the “Claw Plan” alongside GLM-5-Turbo (targeted at agent framework needs). Claw Plan gained 100,000 new users in two days, 400,000 in 20 days. By March 2026, total users on the open platform reached 4 million.
Paid token consumption surged 15-fold from October 2025 to March 2026 (six months).
UBS forecasts open platform revenue to grow from 190 million RMB in 2025 to 6.188 billion RMB in 2027, with a compound growth rate of 470%; total revenue compound growth rate is 231%, from 724 million RMB to 7.941 billion RMB.
The rapid adoption of OpenClaw (a new agent framework) further drives this trend. According to OpenRouter data, as of April 2, 2026, GLM-5-Turbo ranks third in OpenClaw’s token usage.
231% Compound Growth Rate: Revenue Forecast and Risks
UBS expects Zhipu’s total revenue compound annual growth rate from 2025 to 2027 to hit 231%:
- 2025 revenue: 724 million RMB;
- 2026 forecast revenue: 3.208 billion RMB (YOY growth 343%);
- 2027 forecast revenue: 7.941 billion RMB.
Open platform business is forecast to grow at a compound rate of 470%, from 190 million RMB in 2025 to 6.188 billion RMB in 2027, with revenue share rising from 26% to 78%.
Meanwhile, Zhipu remains in a loss-making stage. Estimated net loss in 2026: 5.157 billion RMB; in 2027: 4.747 billion RMB; profitability expected earliest in 2029 (predicted net profit 261 million RMB). As revenue grows, loss rate will gradually shrink.
Main downside risks include: intensified competition in the large model industry; major internet companies (ByteDance, Tencent, Alibaba—all current Zhipu clients) may develop their own models and churn; limits in computing power supply; and geopolitical risks.
Risk warning and disclaimerThe market is risky and investments must be made carefully. This article does not constitute individual investment advice, nor does it consider individual users’ specific investment goals, financial conditions or needs. Users should consider whether any opinions, viewpoints or conclusions herein fit their particular situation. Investment made accordingly is at your own risk.