ByteDance's new AI video tool drove stock gains up to 20%

As big businesses scramble to provide updated products before the approaching Christmas season, ByteDance’s weekend debut of its newest artificial intelligence tool for creating videos sparked new excitement in China’s technology markets.

The business made Seedance 2.0 available on its Jimeng AI platform to a select few users. Testers of the beta version said that, in comparison to previous iterations, the software creates amazingly realistic video material with fluid camera work and improved visual coherence.

Audio advances drive market gains

One standout improvement involves how the system handles audio. Previous models required users to add sound separately after creating videos. The updated version simultaneously creates background noise and conversation to go with the images. Testers emphasized “physics-based realism,” citing precise depictions of motions such as objects falling and character mouth movements that are synced across eight languages.

Market reaction came swiftly on Monday. Huace Media shares climbed roughly 7 percent, while Perfect World gained about 10 percent. COL Group’s stock jumped to its maximum allowed daily increase of 20 percent as investors bet on artificial intelligence boosting traditional media production.

The release comes during what industry watchers describe as a critical period for China’s artificial intelligence industry. Many companies are preparing to launch flagship products.

Source: @alex_prompter

The push to announce new tools before the Lunar New Year reflects fierce worldwide rivalry among leading technology firms competing for user attention during 2026’s opening months, especially after prominent American companies Anthropic and OpenAI made their own major announcements.

Alibaba prepares flagship model release

On Sunday, someone from Alibaba Cloud’s development team submitted pull requests on Hugging Face and GitHub for an upcoming collection of models. Alibaba Cloud handles artificial intelligence and cloud operations for Alibaba Group Holding, which owns the South China Morning Post.

These platforms let programmers share and work together on software code that anyone can access and modify. The centerpiece is Qwen-3.5, arriving nearly a year after Qwen-3’s debut. The earlier version became the world’s most widely used open-model family throughout 2025 thanks to its solid performance, flexible licensing, and broad applications. Information shared through the pull requests indicates Qwen-3.5 will feature two versions, one with 9 billion parameters and another with 35 billion parameters, both offering multimodal capabilities for the first time.

Parameters represent the variables that determine a model’s capabilities and get fine-tuned during development. Higher numbers typically mean stronger performance. Multimodal means the system can work with various data types, such as text, pictures, and sound. Both versions will incorporate the company’s updated architecture, which first appeared in September through an experimental model named Qwen3-Next. Competition has grown more complex with the rise of “agentic” features. Moonshot AI, based in Beijing, recently introduced its Kimi K2.5 model.

This version includes an “agent swarm” function that lets users activate as many as 100 sub-agents for handling multiple tasks simultaneously. The approach follows moves by other startups, including Zhipu AI, which launched GLM-Image, a model allegedly developed using only Chinese-manufactured chips to work around international export limitations.

This year’s “Lunar New Year wave” represents a turning point from the testing phase of artificial intelligence toward widespread integration. Chinese companies are using the holiday season’s heavy internet traffic to draw users into AI-centered platforms.

The aggressive timing demonstrates a developing market where multimodal functionality and open-source availability have become China’s main strategies against American proprietary systems. The emphasis has shifted away from simply building larger models toward specialized agentic abilities and operational efficiency.

China’s AI industry is quickly moving away from experimental models (LLMs) and toward “agentic” ecosystems, which value multi-tasking, real-world processes over raw parameter quantity. In order to circumvent international export restrictions, this generation of flagship releases demonstrates a purposeful emphasis on localized hardware self-sufficiency and open-source dominance.

The smartest crypto minds already read our newsletter. Want in? Join them.