China’s Zhipu AI, MiniMax launch flagship AI models

Chinese startups Zhipu AI and MiniMax have launched new flagship AI models as both prepare for upcoming IPOs in Hong Kong.

Zhipu AI, also known as Z.ai and based in Beijing, introduced its GLM-4.7 model on December 23, highlighting improved coding abilities and high scores on coding benchmarks.

The company said GLM-4.7 achieved a similar milestone to Anthropic’s Claude Opus 4 by surpassing the 70% mark on the SWE-Bench benchmark, which tests real-world software engineering tasks.

Shanghai-based MiniMax released its M2.1 model hours later, reporting improved coding skills in internal tests but still behind Anthropic’s latest model.

Both firms announced their new models shortly after passing regulatory hearings for their Hong Kong listings.

The releases reflect ongoing competition among Chinese AI startups to close the gap with US leaders such as Google DeepMind, Anthropic, and OpenAI.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

Missing financials obscure whether Hong Kong AI model IPOs signal real business traction

Zhipu AI revenue rose from Renminbi (RMB) 57.4 million in 2022 to RMB 312.4 million in 2024. It spent RMB 2.1954 billion on research and development in 2024, about seven times its revenue, which clouds profitability 1.
Zhipu lists 12,000 enterprise customers and 45 million developers. Its CEO put annual recurring revenue above Chinese yuan (CNY) 100 million for developer tools and models, with 2.7 million paying Application Programming Interface (API) users. The filings omit customer concentration, which is reliance on a few clients, and churn rates, which is cancellation pace. GPU (graphics processing unit) infrastructure costs, which is the hardware spend to train and serve models, are missing as well 1 2.
MiniMax reached a US dollars (USD) 2.5 billion valuation after a USD 600 million Series A in March. Backers included Alibaba Group Holding, Tencent Holdings and Sequoia Capital; Hillhouse Group also joined 2.

Asia-Pacific (APAC) enterprises could gain leverage if Chinese AI models offer localized alternatives at competitive pricing

Z.ai calls its GLM Coding Plan “3× usage, 1/7 cost” for enterprise software teams in Asia, but the offer does not name the benchmark; see Z.ai’s pricing page for current rates 3.
Developer tool makers for AI coding assistants can switch traffic between Western and Chinese models to meet compliance needs. GLM-4.6 is an earlier GLM-family model with a 200K context window, the amount of text it can read at once. It came close to Claude Sonnet 4 with a 48.6% win rate on CC-Bench, a code-completion benchmark 4.