China’s model companies are beginning to compete for users in places where software companies once competed for app downloads: startup parks, local events, and developer communities. 36Kr reported on September 10 that Zhipu AI had launched a Hangzhou-focused GLM Coding Plan campaign while Alibaba Cloud’s Qwen Office was preparing token-card distribution for young entrepreneurs in the city. The two initiatives are different in structure, but they point in the same direction. As Chinese AI providers move beyond model launches, they are testing whether direct, local promotion can turn technical capability into repeat usage.
The event is not a new foundational-model release. It is a commercialization and user-acquisition effort. Zhipu’s campaign targets developers and AI-coding users in Hangzhou. Qwen Office’s planned program targets entrepreneurs in university entrepreneurship parks. Both are trying to place AI tools in the hands of people who may incorporate them into workflows, build habits around them, and later become paying or enterprise users.
This approach is notable because China’s AI race has often been described through benchmark scores, parameter counts, and cloud capacity. Those measures remain important, but an AI company also needs a distribution channel. A strong coding model cannot build a durable business if developers do not use it frequently. An office agent cannot become a workplace tool if prospective customers never test it with their own documents and tasks. The Hangzhou campaigns make that commercial problem visible.
Zhipu’s Hangzhou Coding Campaign Targets Developers Directly
36Kr said Zhipu AI launched offline activity tied to its GLM Coding Plan in Hangzhou. A Sina Finance report described the effort as a city-level program involving the Hangzhou municipal government and Shangcheng District, with promotional subscription pricing scheduled to run from September 10 to December 9. The discounts are campaign terms reported by Sina, not a permanent price list or a measure of the service’s general market value.
The campaign’s focus on coding is strategic. Developers can evaluate an AI product repeatedly through code completion, debugging, review, and project work. That creates more opportunities for a company to learn what users need and to convert a free or discounted trial into a subscription. It also exposes the model to a demanding audience that will compare it against both domestic and overseas alternatives.
Zhipu’s product context is already substantial. EastFrontier reported in August that Z.ai launched GLM-5.3 for cybersecurity and coding. The new Hangzhou program does not replace that model announcement. It is a distribution experiment that attempts to move coding capabilities into a concentrated user community. That difference matters because a model release measures technical supply, while a local campaign tests demand and product-market fit.
36Kr also cited marketing metrics after Zhipu opened a store, saying brand search volume rose 50-fold month on month and transactions in AI tokens and related subscription products rose more than 160%. Those figures are outlet-reported promotional indicators, not audited market share or evidence of lasting revenue. They do show why Zhipu may see value in continuing direct outreach: the company has observed a sharp response around a product-access channel.
Qwen Office Uses Token Credits to Reach Startups
Qwen Office’s approach is different. 36Kr said the Hangzhou Talent Management Service Center, working with Qwen Office, was preparing token-card distribution for young people in entrepreneurship parks. Sina reported that the pilot was scheduled to start September 20 across 35 municipal-level university entrepreneurship parks. The program is aimed at early-stage entrepreneurs rather than only professional developers.
The cited report says each card contains 10,000 points with an estimated value of 500 yuan, and that a 10-person startup team could receive up to 5,000 yuan in AI credits. These are promotional terms, not a statement of revenue or a guarantee that recipients will remain customers. But they make the campaign more concrete than a broad invitation to try a product. A startup can apply the credits to real work and decide whether Qwen Office is useful enough to retain after the promotion.
Qwen Office is described in the reporting as an all-in-one platform that can support report writing, presentation creation, data organization, and website building. In other words, it is being marketed as an office assistant rather than a single chatbot. EastFrontier recently reported that QwenWork crossed 30 million users in its first month. The Hangzhou token-card program is a different initiative, but it fits Alibaba’s broader effort to put Qwen-based tools into everyday workplace settings.
The location matters as well. Entrepreneurship parks concentrate new companies that often have limited staff and limited software budgets. They may be more willing to experiment with an AI assistant that promises to reduce time spent on documents, data, or routine web tasks. The announcement does not disclose how Qwen Office will measure success, how many cards it will distribute, or how many users it expects to convert. Those questions will determine whether the program becomes more than a local marketing test.
China’s Model Competition Is Moving From Benchmarks to Habits
The two Hangzhou programs show an emerging shift in Chinese AI competition. Model providers are still releasing new systems and competing on technical performance, but they are also trying to control the path from a model to an active user. In a crowded market, a provider that is easier to access, cheaper to trial, or more visible in a local developer community may gain an advantage even if another company claims a stronger benchmark.
Alibaba’s Qwen has already pursued several distribution routes, including model releases, agent tools, and workplace products. EastFrontier covered Alibaba’s move to take QwenWork beyond China’s workplace AI market. Zhipu has similarly invested in developer-facing models and subscriptions. The September 10 campaigns suggest both companies are now testing offline channels that combine local institutions, credit incentives, and targeted audiences.
That does not mean city-level campaigns will decide China’s model market. Users may switch services, promotions can be expensive, and a free token card does not solve technical issues such as reliability, privacy, integration, or support. The companies will need to show that recipients continue using their tools after the initial incentive ends. They will also need to prove that usage can translate into revenue without sacrificing the low prices that helped attract users.
For now, the Hangzhou initiatives provide a clear sign of market maturation. China’s leading AI providers are no longer only competing to announce the next model. They are competing to become part of a developer’s daily coding routine or an entrepreneur’s office workflow. The winner may not be the company with the loudest launch. It may be the one that can turn a local promotion into lasting product habit.