China AI Labs Race to Launch Models Before Lunar New Year

Chinese artificial intelligence laboratories are accelerating the release of new generative models in the final weeks of the Year of the Snake, aiming to debut their latest systems before the Lunar New Year holiday commencing February 15, 2026, according to South China Morning Post reports.

Beijing-based Zhipu AI (known internationally as Z.ai) plans to launch the fifth iteration of its GLM model series, GLM-5, featuring significant enhancements in creative writing, coding, reasoning, and agentic capabilities, with the release expected within two weeks of the report. Shanghai-headquartered MiniMax, which recently completed its Hong Kong stock exchange debut, is preparing to release an updated M2.2 model focused on coding improvements, according to sources familiar with the companies' plans.

While Zhipu and MiniMax declined to comment on their release schedules, the timing aligns with an established industry practice where Chinese AI firms leverage the Lunar New Year holiday—a period when most Chinese citizens take extended leave—to maximize product visibility and user acquisition. The festival provides a captive audience for new model launches, as noted in a Global Semi Research analysis, which highlighted that ByteDance's Seedance 2.0 and Alibaba's Qwen 3.5 are also scheduled for release during this window.

Alibaba's Qwen 3.5 is anticipated to adopt a novel hybrid attention mechanism and function as a native vision-language model, while the company has already unveiled two supporting models: Qwen-Image-2.0 for professional image generation and RynnBrain for robotics intelligence. Alibaba is also reportedly spending 3 billion yuan on promotional incentives for its Qwen chatbot, as detailed in the Substack analysis. Meanwhile, DeepSeek is preparing its V4 model with integrated long and short chain-of-thought reasoning capabilities, according to the same report.

Zhipu's GLM-5 development is particularly notable for its reliance on domestic hardware, with the company training the model on a 100,000-chip Huawei Ascend cluster using a fully Chinese hardware pipeline, as reported by Digital Applied. This represents a strategic shift toward hardware independence, following Zhipu's earlier GLM-Image model trained on Huawei's Atlas 800T A2 hardware using the MindSpore AI framework. The 2026 Lunar New Year period is expected to feature five major model releases across five distinct companies, marking what analysts describe as the most concentrated Chinese AI model launch cycle to date.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn