Chinese AI models festoon Spring Festival a year after DeepSeek shock
(Updates Feb 14 story to add Alibaba’s Qwen3.5 release and Zhipu’s Monday statement)
BEIJING, Feb 18 (Reuters) – As China celebrates the Lunar New Year holiday, rivals to DeepSeek have scrambled to release artificial-intelligence models a year after it burst onto the scene with its game-changing R1 and V3 models. With DeepSeek set to launch its next-generation V4 model soon according to tech news site The Information, many other Chinese AI firms have released their own models in the hopes of stealing the spotlight – or at least avoiding being caught off-guard again – during this year’s Spring Festival.
Below are the companies and models that made, or could make, a Spring Festival splash:
DEEPSEEK The Hangzhou-based startup’s V4 would replace last year’s V3 model, which powered the AI assistant app that overtook ChatGPT to become the top-rated free application available on Apple’s App Store in the U.S. Investors and industry insiders are also on the lookout for R2, successor to the R1 model.
Last week, DeepSeek fuelled anticipation when its web and mobile chatbot upgraded its “context window” – the amount of information it can remember and handle in a single task, from 128,000 to 1 million tokens, the unit of data processed by the AI model.
This means the chatbot can now process book-length passages of text to answer a single user command.
BYTEDANCE The makers of short-video platform TikTok released the Doubao 2.0 chatbot, the company said on Saturday. The new model is capable of deep thinking and long, multi-step task execution that matches the capabilities of U.S. rival OpenAI’s GPT 5.2 and Google’s Gemini 3 Pro, Bytedance said.
Doubao is China’s most popular AI chatbot app, in terms of weekly active users, according to data published by Quest Mobile in December. Thursday’s release of video-generation AI model Seedance 2.0 has generated comparisons to DeepSeek’s global rise, going viral on Chinese social media and drawing widespread praise on X, including from the platform’s owner, Elon Musk. Seedance 2.0 can produce high-quality cinematic videos based on a few prompts, or even one.
The tech giant released picture-generation model Seedream 5.0 Lite on Friday.
ALIBABA Alibaba, the first Chinese firm to respond to DeepSeek’s viral ascent last year with its Qwen 2.5-Max, released Qwen3.5 on Monday, a model the company described as being “built for the agentic AI era”.
The focus on agents comes as expectations grow among consumers, developers and enterprises that AI can go beyond the chatbot framework to execute a much wider range of tasks, from shopping on different apps to managing complex workflows with minimal human interference. The e-commerce giant’s Qwen app is riding a wave of growing domestic usage after it spent 3 billion yuan ($400 million) last week on a coupon giveaway campaign to promote “agentic commerce”, where AI handles consumers’ online shopping.
This drove more than 120 million consumer orders in the six days through Wednesday, the company said.
ZHIPU Zhipu AI released its open-source GLM-5 model last week, with enhanced coding capabilities and the ability to perform long-running agent tasks.
Zhipu said on Monday it saw an instant spike in global usage following GLM-5’s release, exceeding existing capacity and leading to delays for some users. The company added that it had expanded its domestic chip cluster multiple times and introduced limited paid plans but this had not fully resolved the issue, prompting it to invite companies with chip clusters to collaborate on optimising GLM-5 at the hardware and systems level.
Zhipu is considered one of China’s “AI tigers” – promising startups vying with the U.S. to win the AI race. Zhipu went public on the Hong Kong Stock Exchange last month, alongside rival MiniMax, another AI tiger. Both stocks have rallied strongly as investors bet on the companies benefiting from China’s AI boom. Zhipu plans a secondary listing in Shanghai, a regulatory filing last week showed.
MINIMAX MiniMax released its M2.5 open-source model last week. The company’s Hong Kong listing raised HK$4.8 billion ($620 million), higher than Zhipu’s $558 million.
Shanghai-based MiniMax has developed popular apps like Hailuo AI, a video generation tool, and Talkie, a character interaction app that enables users to engage with AI-powered virtual personas.
TENCENT Tencent’s Hunyuan team last week released a low-storage, compressed AI model, HY-1.8B-2Bit, designed to be used on consumer hardware including mobile phones.
IFLYTEK iFlytek last week released Spark X2, trained entirely on Chinese-made chips. The company said the upgrade focuses on practical deployment in sectors including education, healthcare, automotive and agent-based applications.
NETEASE YOUDAO NetEase Youdao last week launched LobsterAI, a desktop-level personal assistant agent that can perform tasks such as information retrieval, scheduling and data analysis by executing workflows locally on a user’s computer after authorisation.
The product supports mobile and PC connections and allows remote interaction via enterprise apps popular among Chinese companies such as DingTalk and Feishu.
DEXMAL Embodied-intelligence startup Dexmal last week unveiled DM0, an AI model designed for robot-related scenarios. DM0 integrates multimodal internet data with driving, navigation and robotic operation data, and was trained across multiple robot platforms. (Reporting by Eduardo Baptista; Editing by William Mallard)
Post Comment