Chinese AI Models Flood the Spring Festival Spotlight One Year After the DeepSeek Shock

BEIJING As China gears up for the Lunar New Year holiday, the country’s artificial intelligence sector is buzzing with activity. A year after DeepSeek stunned the industry with its breakthrough R1 and V3 models, rivals are racing to unveil new AI systems — determined not to be overshadowed again during this year’s Spring Festival.

Last year, DeepSeek’s surprise momentum caught competitors off guard. This time, China’s tech giants and rising AI startups are launching upgrades and entirely new models in rapid succession, hoping to capture attention before DeepSeek’s anticipated next move.

DeepSeek: Raising the Bar Again

Hangzhou-based DeepSeek is reportedly preparing to release its next-generation V4 model, which would replace the widely adopted V3 system that powered its AI assistant app. That app famously surpassed OpenAI’s ChatGPT to become the top-rated free app on Apple’s U.S. App Store — a symbolic victory for China’s AI ambitions.

Investors are also watching for the successor to its R1 model, likely to be called R2.

Adding to the anticipation, DeepSeek recently expanded its chatbot’s “context window” — the amount of information it can process in a single interaction — from 128,000 tokens to an impressive 1 million tokens. The upgrade signals the company’s push into more complex, multi-step reasoning tasks.

ByteDance: Doubao and the Viral Video Push

TikTok parent ByteDance has rolled out Doubao 2.0, an upgraded chatbot designed for deeper reasoning and longer task execution. The company claims its performance rivals GPT-5.2 and Gemini 3 Pro.

Doubao is already China’s most popular AI chatbot in terms of weekly active users, according to industry data.

ByteDance also grabbed attention with Seedance 2.0, a video-generation AI model that quickly went viral on Chinese social media and gained praise on X — including from platform owner Elon Musk. The model can generate cinematic-quality videos from just a few prompts, sometimes even a single line.

The company followed up with Seedream 5.0 Lite, a new image-generation model, further strengthening its multimodal AI portfolio.

Alibaba: Betting Big on Agentic Commerce

E-commerce giant Alibaba, one of the first major firms to counter DeepSeek’s rise last year with Qwen 2.5-Max, is now preparing to launch Qwen 3.5.

Alibaba’s strategy goes beyond model releases. The company recently invested 3 billion yuan (about $400 million) in a coupon campaign promoting “agentic commerce,” where AI systems manage online shopping on behalf of users. The push reportedly drove over 120 million consumer orders within six days — highlighting how AI is becoming embedded directly into commercial activity.

Zhipu and MiniMax: The “AI Tigers”

Startup Zhipu AI has introduced its open-source GLM-5 model, featuring stronger coding abilities and support for long-running agent tasks. Zhipu is often described as one of China’s “AI tigers” — a group of promising startups competing with U.S. leaders in advanced AI.

Zhipu recently went public in Hong Kong and is reportedly planning a secondary listing in Shanghai.

Meanwhile, Shanghai-based MiniMax launched its M2.5 open-source model on its overseas agent platform. Its Hong Kong IPO raised more capital than Zhipu’s, reflecting strong investor enthusiasm.

MiniMax is also known for consumer-facing AI products such as Hailuo AI, a video generation tool, and Talkie, a character interaction app featuring AI-powered virtual personas.

Tencent, iFlytek, and Others Join the Race

Tech heavyweight Tencent released a compressed AI model, HY-1.8B-2Bit, designed to run efficiently on consumer devices, including smartphones — signaling a push toward on-device AI.

iFlytek unveiled Spark X2, trained entirely on Chinese-made chips. The model is focused on real-world deployment in sectors such as education, healthcare, and automotive systems.

NetEase’s Youdao unit launched LobsterAI, a desktop-level personal assistant capable of executing workflows locally on users’ computers. It also integrates with widely used enterprise platforms in China.

Embodied-intelligence startup Dexmal entered the spotlight with DM0, a model built for robotics applications. By combining multimodal internet data with driving and navigation inputs, DM0 is designed for real-world robotic operations across multiple platforms.

A High-Stakes AI Spring Festival

The flurry of announcements reflects more than holiday timing. China’s AI sector is entering a new competitive phase — one defined by longer context windows, multimodal capabilities, agent-based workflows, robotics integration, and commercial deployment.

A year ago, DeepSeek caught the market by surprise. This Spring Festival, no one wants to be left unprepared.

As anticipation builds for DeepSeek’s V4 — and possibly R2 — China’s AI landscape is no longer reacting. It’s accelerating.