Chinese AI developers have treated this year’s Spring Festival period as a strategic window to release or upgrade models, after the shock entry of Hangzhou-based DeepSeek last year with its R1 and V3 systems. With reports that DeepSeek is poised to introduce a successor V4 model soon, other domestic players moved quickly to avoid being eclipsed again - and to compete for user attention during the holiday.
The wave of announcements covers chatbots, multimodal image and video generators, agent-focused systems designed to execute complex tasks, and models aimed at robotics applications. Below is a company-by-company account of the recent launches and the market or operational developments they produced.
DeepSeek
DeepSeek, whose V3 model last year powered an AI assistant app that overtook ChatGPT to become the top-rated free app on Apple’s U.S. App Store, is expected to follow up with a V4 release, according to tech news site The Information. The startup has also stirred interest around R2, the intended successor to its R1 family. In a high-profile upgrade last week, DeepSeek expanded the chatbot’s "context window" - the amount of information it can retain and process in a single task - from 128,000 to 1 million tokens. That capacity increase allows the assistant to handle book-length passages and respond to single commands that require processing very large volumes of text.
Bytedance
ByteDance released Doubao 2.0, a next-generation chatbot it said can undertake deep reasoning and long, multi-step tasks. The company positioned Doubao 2.0 as competitive with earlier-generation models from U.S. rivals, asserting parity with named alternatives. Doubao remains China’s most-used chatbot app by weekly active users, according to data cited in December.
In addition to chat, ByteDance has pushed video and image-generation models into the spotlight. The company launched Seedance 2.0, a video-generation model that can yield cinematic-quality videos from just a few prompts or even a single prompt. Seedance 2.0 quickly spread across Chinese social platforms and drew wide praise on X, including from that platform’s owner, Elon Musk. ByteDance also released Seedream 5.0 Lite, a picture-generation model, during the same period.
Alibaba
Alibaba, which earlier responded to DeepSeek’s rise with its Qwen 2.5-Max, introduced Qwen3.5 this week. The company described the new model as "built for the agentic AI era," reflecting a focus on so-called agents - systems designed to perform a broader set of tasks autonomously, from shopping across multiple apps to coordinating workflows with limited human intervention.
The e-commerce giant has moved to promote agent-based commerce broadly. Alibaba said it spent 3 billion yuan ($400 million) on a coupon giveaway to drive adoption of agentic shopping features. The company reported that the coupon campaign generated more than 120 million consumer orders in the six days through Wednesday, and its Qwen app is seeing rising domestic usage as a result.
Zhipu
Zhipu AI released the open-source GLM-5 model last week, emphasizing stronger coding skills and the ability to sustain long-running agent tasks. The company reported an immediate surge in global usage after GLM-5’s launch, which exceeded available capacity and produced delays for some users. Zhipu said it has repeatedly expanded its domestic chip cluster and introduced limited paid plans, but those steps did not fully resolve strain on its systems. The firm invited companies with large chip clusters to collaborate on hardware and systems optimisation for GLM-5.
Zhipu is commonly counted among China’s so-called "AI tigers" - startups that investors view as high-potential challengers in the global AI race. The company completed a public listing on the Hong Kong Stock Exchange last month, and earlier this year its shares joined a broader rally in AI-related stocks. A regulatory filing last week showed plans by Zhipu for a secondary listing in Shanghai.
MiniMax
MiniMax followed with the release of an open-source M2.5 model last week. The Shanghai-headquartered company raised HK$4.8 billion ($620 million) in its Hong Kong listing, a larger sum than the roughly $558 million raised by Zhipu in its offering. MiniMax has developed consumer-facing applications such as Hailuo AI, a video-generation tool, and Talkie, which enables interactions with AI-driven virtual characters.
Tencent
Tencent’s Hunyuan team introduced a compressed, low-storage model named HY-1.8B-2Bit. The model is designed to run efficiently on consumer-grade hardware, including mobile phones, by reducing storage requirements and enabling broader on-device deployment.
iFlytek
iFlytek upgraded its offerings with Spark X2, a model trained entirely on domestically produced chips. The company said the update is aimed at practical deployments across education, healthcare, automotive and agent-based use cases.
NetEase Youdao
NetEase Youdao launched LobsterAI, which the firm describes as a desktop-level personal assistant agent. LobsterAI can retrieve information, schedule tasks and perform data analysis by executing authorisation-approved workflows locally on a user’s computer. The product supports mobile and PC connectivity and allows remote interaction via enterprise apps widely used in China, such as DingTalk and Feishu.
Dexmal
Embodied-intelligence startup Dexmal unveiled DM0, an AI model tailored for robot scenarios. DM0 combines multimodal internet-sourced data with navigation, driving and robotic operation datasets and was trained across multiple robot platforms to support practical robotic control tasks.
Across the board, model rollouts have been accompanied by a mixture of product celebration, consumer attention and operational stress. Some companies reported capacity challenges and have sought hardware partnerships or paid-plan rollouts to manage demand. Others have tied model launches to marketing and transaction-promoting campaigns designed to drive app engagement and commerce.
For investors, the period has seen market reactions to listings and product announcements - most notably the recent Hong Kong listings and the subsequent rallies in AI-related stocks. Firms that combined high-profile model launches with capital markets activity drew particular attention from backers betting on growth in AI adoption.
The pace and variety of model introductions during the Spring Festival underscore how China’s AI sector is positioning across multiple fronts - consumer chat and media generation, enterprise agent workflows, device-level models and robotics. The outcomes for usage patterns, infrastructure needs and corporate revenues will be visible only as deployments scale and users adopt these newly released capabilities.