Business judgment on AI products
There is public usage data. The useful view is what it does, its scale, and whether the change lasts.
01
Start inside the user's day · Public facts + observable behavior · 2026-08-28
Ordinary users open 智谱清言 and talk to GLM. Monthly visits: 4.7 million.
Public materials do not yet show how users complete this job today or what they replace.
The product targets friction in this job, but public user evidence does not yet show the cost, frequency, or consequence of leaving it unsolved.
A public record shows 4.69M monthly visits and 6.66% month-over-month growth. That explains the attention, but product-level retention and payment are not yet verified.
① Whether web visits keep positive MoM — a scarce signal in a topping-out market; ② Whether the "50M MAU" media figure gets confirmed by financials — the two; series differ by 10x and must be reconciled; ③ Whether Qingyan traffic rises in step with the next GLM release — testing the; "model pulls th…
Worth dissecting. A public record shows 4.69M monthly visits and 6.66% month-over-month growth. That explains the attention, but product-level retention and payment are not yet verified.
a model company building consumer product should not expect the entry point to make money by itself. Make it a demo window plus an enterprise-lead feeder; C-side traffic feeds B-side contracts. The product's job is "let people try the model," not "produce revenue." the same model split into different paid products by work scenario (Coding Plan, Claw Plan). One base model, priced by what the user does with it, captures willingness-to-pay better than a single subscription. This splitting approach transfers to other businesses.
Free chat plus membership subscription; The real revenue lines: MaaS/API calls, enterprise solutions, GLM Coding Plan; 2025 full-year revenue ¥724M (+131.9% YoY); MaaS platform ARR ~¥1.7B, up 60x in 12 months; July 2026: ~HKD 31.4B placemen… ① Whether web visits keep positive MoM — a scarce signal in a topping-out market; ② Whether the "50M MAU" media figure gets confirmed by financials — the two; series differ by 10x and must be reconciled; ③ Whether Qingyan traffic rises in step with the next GLM release — testing the; "model pulls th…
Ordinary users open 智谱清言 and talk to GLM. Monthly visits: 4.7 million.
A public record shows 4.69M monthly visits and 6.66% month-over-month growth. That explains the attention, but product-level retention and payment are not yet verified.
An English validation note will follow from the public evidence.
The product claims to help users complete: “Ordinary users open 智谱清言 and talk to GLM. Monthly visits: 4.7 million.”. User evidence has not yet verified pain intensity or the cost of doing without it.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
The assessment is recorded; an English explanation is pending.
02
Market comparison
The Chinese–English market comparison is not complete yet. A conclusion follows only after its coverage and verifiable evidence are recorded.
03
The call and next move come first; the full read retains the evidence and counterevidence.
Zhipu's consumer chat entry point, sitting on the GLM models — but the company's money comes mainly from enterprise MaaS and corporate customers. Qingyan is the storefront, not the granary.
Zhipu (Beijing Zhipu Huazhang Technology), one of China's earliest independent model labs, listed on the HKEX on 2026-01-08 (offer price HKD 116.2), billed as the "world's first listed AGI foundation-model company." Qingyan is its consumer app/web product on the same technical line as the GLM base models.
Read: understand Qingyan only in the context of Zhipu's business structure. The company's center of gravity has never been consumer apps; Qingyan reads as a showcase window and brand front door for GLM capability.
What it deliberately does not do: it does not own a vertical scenario. Qingyan is a general-purpose chat entry, sitting on the same shelf as Doubao, Kimi, and Yuanbao.
Previously, "using GLM" had only two paths: writing code to call the API, or waiting for some company to integrate the model into its own product. Qingyan gave ordinary users a third path — download an app and just talk. It is the shortest route from GLM capability to the consumer.
For users, it replaces the generic act of "chatting with a model in a browser or terminal." For the industry, it replaces the high-cost path of "deploying an open-source model yourself" with pay-as-you-go subscriptions and APIs.
Read: Qingyan membership is a rounding error in Zhipu's revenue map. That sets the product strategy — Qingyan does not need to monetize like Doubao. Its job is to make GLM seen, tried, and noticed by enterprise buyers.
| Dimension | Call |
|---|---|
| Founder-product fit | A corporate project; founders (Zhang Peng / Tang Jie's team) are academic-model people |
| Product insight | No differentiating killer move as a chat product; the strengths are all model-side |
| Execution quality | GLM series is first-tier domestically, strong at agentic/coding tasks |
| Timing | Enterprise MaaS and coding subscriptions are booming; consumer chat is a red ocean |
Qingyan is a live specimen of "strong model, weak entry point." Zhipu validates a long-standing read: Chinese model companies are consistently weaker on product than on models and APIs. Qingyan's traffic is an order of magnitude below Doubao (~320M), DeepSeek, or Kimi — yet the company's API volume kept climbing even after a 30% price hike. Customers buy the model, not the Qingyan app.
+6.66% MoM on the web is the most interesting signal in this data set. In 2026, when consumer chat has largely topped out, Qingyan's web side is still climbing gently — real organic growth, just with a ceiling well below traffic giants like Doubao.
Its industry position is worth noting: when a model company's consumer app sells less well than its API, the market has accepted "model as a service" rather than "chat as an entry." Qingyan therefore does not need to be #1. Good enough, trustworthy, always available — that is sufficient to feed leads and endorsement to the parent's enterprise sales.
Risk: if Qingyan remains only an entry, it has no standalone commercial story. Should Zhipu's strategic weight shift further toward enterprise, investment in Qingyan may be downgraded into a pure demo product.
① Whether web visits keep positive MoM — a scarce signal in a topping-out market ② Whether the "50M MAU" media figure gets confirmed by financials — the two series differ by 10x and must be reconciled ③ Whether Qingyan traffic rises in step with the next GLM release — testing the "model pulls the entry point" causal chain
Product logic: a model company building consumer product should not expect the entry point to make money by itself. Make it a demo window plus an enterprise-lead feeder; C-side traffic feeds B-side contracts. The product's job is "let people try the model," not "produce revenue."
Pricing structure: the same model split into different paid products by work scenario (Coding Plan, Claw Plan). One base model, priced by what the user does with it, captures willingness-to-pay better than a single subscription. This splitting approach transfers to other businesses.
Worth watching. The company's fundamentals are solid and public; Qingyan, as GLM's consumer window, grows mildly without stalling. The real thing to watch is not Qingyan itself but the validated "model as a service" path — and how long Qingyan's web traffic keeps climbing.
05
Use these searches when the official site is missing or the current link is only a lead.