Z.ai offers up to 5 trillion GLM-5.3 tokens to recruit ZCode users

Why it matters

Z.ai is using model capacity as customer-acquisition spend. The campaign tests whether a large free allowance can turn GLM-5.3 interest into durable ZCode usage while its serving infrastructure handles the load.

Z.ai, the Tsinghua-founded AI company led by chairman and co-founder Liu Debing and chief executive and co-founder Zhang Peng, is offering 100 million free GLM-5.3 tokens apiece to 50,000 new users through August 23 at 6 p.m. Pacific.

The maximum advertised allocation is 5 trillion tokens. Z.ai said it was extending its Build Week promotion into an ongoing series after users built projects with ZCode and GLM-5.3. The allocation remains a Z.ai claim, and tokens are a unit of model usage rather than a measure of Z.ai's cash cost.

Z.ai on X

The promotion gives Z.ai two days to convert interest in a week-old model release into new ZCode users. ZCode is a desktop coding agent. It can plan changes, edit code, run tools and manage longer tasks, and its current product page promotes deep GLM-5.2 integration across reasoning, coding and multi-agent collaboration. The current promotion separately pairs ZCode with GLM-5.3. ZCode also lets developers steer jobs remotely through WeChat, Feishu or Telegram.

The Tsinghua founders are buying distribution with compute

Z.ai grew out of research at Tsinghua University rather than the consumer-software circles that produced many of the leading Western coding tools. Its Hong Kong listing prospectus identifies Liu, Zhang, Tang Jie, Li Juanzi and Xu Bin as co-founders. Liu, a former Technicolor research engineer and Tsinghua senior engineer, serves as chairman. Zhang, who worked at Tsinghua from 2005 through 2020, is chief executive.

Tang and Li supply much of the academic lineage behind the company. Tang created AMiner, a search and network-analysis platform for scientific research, and later led work on the Wu Dao large-model project. Li, a Tsinghua professor, directs the university's Knowledge Intelligence Center and has spent her career working on knowledge graphs, data mining and language processing.

That background shaped Z.ai's original approach: publish models, build a developer base around them and sell the infrastructure needed to put those models into production. Z.ai's own history traces the company to Tsinghua technology and records the 2022 release of GLM-130B, an early open model at the 100-billion-parameter scale.

ZCode extends that playbook into a market where the application can matter as much as the underlying model. Cursor, Claude Code and GitHub Copilot already sit inside developers' daily workflows. A benchmark gain may persuade engineers to test GLM-5.3. A free allowance large enough to support sustained work gives Z.ai a better chance of making ZCode the place where they keep using it.

GLM-5.3 moved from paper to promotion in one week

The giveaway caps a compressed product cycle. Z.ai previously introduced ZCode as a free desktop application designed to turn GLM-5.2 into a distribution channel for its paid coding plans. On August 11, Z.ai said ZCode had reached 1 million users, though that self-reported figure did not distinguish registered accounts from active or paying developers.

Z.ai announced GLM-5.3 on August 14, 2026, initially without a documented access path. Z.ai later opened GLM-5.3 through its API. The current token campaign turns that technical release into a customer-acquisition push.

In its GLM-5.3 launch post, Z.ai said the model uses the same base as GLM-5.2 with additional post-training. The company claims a 50% improvement on its internal coding benchmark. It also reported that GLM-5.3 raised its score from 4.6 to 28.3 on Terminal-Bench 3.0 and from 46.2 to 66.9 on DeepSWE v1.1. Those figures come from Z.ai's evaluation and have not been independently reproduced here.

The commercial logic is straightforward. Coding agents consume substantial inference capacity because they repeatedly inspect repositories, generate changes, run commands and revise their work. A 100 million-token allowance gives a new user room to attempt longer projects instead of judging ZCode from a handful of prompts. It also steers the user toward Z.ai's own interface, where Z.ai controls onboarding, model selection and the path to a paid plan.

Five trillion tokens will also test the serving layer

The campaign creates an infrastructure test alongside the distribution opportunity. If all 50,000 allocations are claimed and heavily used, Z.ai must serve a theoretical maximum of 5 trillion tokens while keeping latency and reliability acceptable for coding work.

Z.ai has previously acknowledged scaling problems affecting GLM-5-series coding agents during periods of heavy demand, including repetition and garbled output. The free-token pool can expose ZCode to thousands of prospective users, while any recurrence of those problems would shape their first experience with the product.

Z.ai has the capital to subsidize adoption. The company raised approximately HK$4 billion, or about $558 million, in a Hong Kong IPO in January 2026 at HK$116.20 per share, according to its prospectus. The filing also showed why management remains focused on growth: Z.ai recorded RMB190.9 million in revenue and a RMB2.36 billion loss during the first half of 2025, with research and development accounting for much of its spending.

The giveaway should not be read as a 5-trillion-token cash expense. Z.ai's actual cost depends on how many users qualify, how much of each allowance they consume and the efficiency of the infrastructure serving GLM-5.3. The useful number comes later: how many developers are still opening ZCode once the free allocation is gone.