Call volume increased by about 5 times, price is only one-tenth of Claude, and the market cap of Zhipu and MiniMax both surpassed 300 billion HKD. What is the market rushing to buy?

February 20th, the first trading day of the Year of the Horse in the Hong Kong stock market, saw Zhipu, known as the “world’s top large model stock,” close up over 42%, with a market capitalization surpassing HKD 323.2 billion. Since February, the company’s stock price has increased by over 220%, and since its listing, the increase has reached 523%.

Another AI large model company, MiniMax, continued its strong momentum before the holiday, closing up over 14%, with a market value exceeding HKD 300 billion. Since listing, MiniMax has accumulated a total increase of 487.88%.

Currently, both companies’ market caps have surpassed Ctrip and Kuaishou, approaching Pop Mart (HKD 327.9 billion) and Baidu (HKD 354.8 billion).

However, amid the frenzy, both companies are still operating at a loss, with a price-to-sales ratio of up to 700 times, far exceeding OpenAI’s 65 times. What kind of future is the market betting on?

Zhipu and MiniMax both surpass HKD 300 billion in market cap, with PS ratios far exceeding OpenAI

On February 20th, the first trading day of the Year of the Horse according to the lunar calendar, while the Hang Seng Tech Index fell nearly 3% and many traditional tech stocks underperformed, two AI large model startups—Zhipu and MiniMax—became among the few bright spots in the market.

That day, Zhipu closed at HKD 725 per share, up 42.72%, with a market cap exceeding HKD 323.2 billion, a single-day increase of over HKD 96.7 billion. MiniMax also performed well, closing up 14.52% at HKD 970 per share, with a market cap of HKD 304.2 billion, up over 105% in February.

Notably, these two companies have not been listed long. Zhipu went public on January 8, 2026, with an issue price of HKD 116.20 per share; MiniMax followed shortly after, listing on January 9 at HKD 165 per share. In just over a month, Zhipu’s cumulative increase has reached 523%, and MiniMax’s has hit 487.88%.

Currently, Zhipu and MiniMax have surpassed short-video platform Kuaishou (HKD 289.4 billion) and travel giant Ctrip (HKD 286.7 billion), and are close to the market caps of Pop Mart (HKD 327.9 billion) and search giant Baidu (HKD 354.8 billion).

Hu Yanping, a distinguished professor at Shanghai University of Finance and Economics, told the Daily Economic News (hereafter DEN) that “these two companies are somewhat overvalued compared to mature companies like JD.com, but their valuations are not high compared to US AI companies. The market’s trust and valuation expectations for ‘small tigers’ outside the big firms have increased.”

Despite the enthusiasm in the secondary market, both companies are still deep in losses. According to prospectuses and financial reports, from 2022 to the first half of 2025, Zhipu’s accumulated losses totaled 6.238 billion yuan. While MiniMax’s revenue in the first nine months of 2025 grew 174.76% year-over-year to $53.44 million, its net loss reached $512 million (about RMB 3.605 billion).

Both companies’ PS ratios exceed 700 times, far higher than international giants like OpenAI. Based on the recent valuation of $850 billion and projected revenue of $13 billion in 2025, OpenAI’s PS ratio is 65 times.

Zhipu GLM-5 and MiniMax M2.5 ignite the market, with weekly call volume increasing over 460%

Behind the market enthusiasm are key technological and product breakthroughs by these two companies.

On February 12th, Zhipu officially open-sourced its new flagship model GLM-5, marking the shift from the “code snippet writing” vibe coding era to the “system engineering completion” intelligent agent engineering era.

GLM-5’s performance in programming scenarios improved by over 20% compared to the previous generation, with real coding experience approaching Claude Opus 4.5 level. Additionally, in evaluations such as BrowseComp, MCP-Atlas, and τ2-Bench, GLM-5 achieved top performance in the open-source domain.

Following the release of GLM-5, due to high demand, Zhipu increased the price of its GLM Coding Plan packages the next day—by 30% in China and over 100% overseas—becoming the first domestic AI native company to raise prices for large model commercialization services. The new packages sold out immediately, setting a new industry record for paid Chinese AI programming models.

On the same day, MiniMax released M2.5, a production-level model designed for agent scenarios, supporting full-stack programming development across PC, app, and cross-platform applications, targeting high-frequency productivity scenarios such as coding, tool invocation, search, and office work.

In public benchmark tests, M2.5 achieved or surpassed industry SOTA levels in multiple core metrics, including SWE-Bench Verified (80.2%), Multi-SWE-Bench (51.3%), and BrowseComp (76.3%). Compared to the previous M2.1, M2.5 shows significant improvements in task decomposition efficiency and token consumption during reasoning, completing tasks 37% faster in SWE-Bench Verified, making it more suitable for long-chain, continuous agentic tasks.

In the recently popular open-source AI agent project OpenClaw, MiniMax’s model is among the most favored choices. Its founder stated that “for programming tasks, MiniMax can achieve near-top closed-source model performance at about 5% of the cost.”

Data from the model routing platform OpenRouter shows that MiniMax M2.5 and Zhipu GLM-5 have significant cost advantages over Claude Opus 4.6: in input costs, both are priced at $0.3 per million tokens, while Claude Opus 4.6 costs $5 per million tokens—about 16.7 times higher; in output costs, MiniMax M2.5 is $1.1 per million tokens, GLM-5 is $2.55, and Claude Opus 4.6 is $25, roughly 22.7 and 9.8 times higher respectively.

OpenRouter’s latest data also shows that MiniMax M2.5 is the most called model this week, with 3.07 trillion tokens used, a 524% increase week-over-week; GLM-5 ranks third with 1.03 trillion tokens, up 462%.

Token demand “inflation” becomes “production fuel”

Behind the market frenzy is the harsh reality that both companies are still operating at huge losses.

According to prospectuses and financial reports, from 2022 to the first half of 2025, Zhipu’s total losses exceeded 6.2 billion yuan. MiniMax, despite a 174.7% year-over-year revenue growth to $53.44 million in the first nine months of 2025, also reported a net loss of $512 million (about RMB 3.605 billion).

Why is the market willing to assign such high valuations?

Hu Yanping analyzed for DEN that, besides increased trust and valuation expectations for “small tigers” outside the big firms, it also relates to the absence of a listed company like Kimi. Big tech firms’ businesses are mixed with other segments, leading to unclear valuation paths for Chinese AI. “In fact, companies like Kimi and DeepSeek are generally better than these two already listed companies.”

He further emphasized, “Whether they have formed a closed loop across four dimensions—technology R&D, product services, application markets, and capital markets—is very important. From this perspective, these two (Zhipu and MiniMax) have ‘broken out.’ Their recent products in coding, multimodal, and agent support perform well, and some aspects even rival big firms’ products, with faster monthly updates and iterations.”

Analysis from Guolian Minsheng Securities points out that in the large model era, tokens are not just “traffic” with near-zero marginal cost but essential “fuel” for executing production tasks. Zhipu’s price increase indicates a shift in the measurement unit from traffic (DAU/duration) to tokens (inference consumption), which are increasingly a necessity in many scenarios.

As AI roles evolve from simple Q&A to actual “work”—such as code reconstruction, document generation, and testing—token consumption will surge. This shift is further amplified by the rise of agents, which can proactively plan, retrieve, execute, and reflect. A complex task may require dozens or hundreds of model calls, each adding to token consumption. The market expects that in the future, a user might run multiple agents simultaneously on different tasks, with daily token consumption reaching tens of millions.

Guolian Minsheng Securities states that when inference consumption becomes a production resource, large model vendors can turn “computing power scarcity” into profits and cash flow through tiered pricing and subscription products. If they can maintain subscription retention and enterprise expansion in high-ROI scenarios like programming, agents, and enterprise workflows, and convert “token usage” into delivery value that saves labor, time, and rework, they will have the ability to transcend open-source and price wars.

However, this also involves significant uncertainties.

First, the business models of these two companies still need stabilization. Over 70% of MiniMax’s revenue comes from overseas, and its main product Talkie’s user stickiness and paid conversion rate abroad have yet to be proven long-term, making it sensitive to external changes. Zhipu’s programming models also face fierce competition from top global products like Claude, Gemini, and GPT, and how long it can maintain pricing power remains uncertain.

Second, the scarcity of AI stocks in Hong Kong amplifies valuation premiums. As one of the few pure large model stocks in the Hong Kong market, Zhipu and MiniMax attract a large influx of funds seeking AI investment exits. But compared to strategic investors like Microsoft and Amazon backing OpenAI, their shareholder structures and ecosystem synergies still lag behind.

Disclaimer: The content and data in this article are for reference only and do not constitute investment advice. Please verify before acting. Use at your own risk.

Daily Economic News