You're offline - Playing from downloaded podcasts
Back to All Episodes
Podcast Episode

Chinese AI Giants Unleash Wave of New Models Ahead of Lunar New Year

February 15, 2026

0:00
4:06
Podcast Thumbnail

China's biggest tech companies are releasing a flurry of advanced AI models spanning video generation, robotics, and large language models ahead of the Spring Festival. ByteDance, Kuaishou, Zhipu, MiniMax, and Alibaba have all unveiled major products while spending hundreds of millions on user promotions.

A Year After DeepSeek, China's AI Sector Enters a New Phase

Exactly one year after DeepSeek's breakthrough R1 reasoning model sent shockwaves through the global AI industry, China's tech giants are marking the anniversary with an unprecedented burst of new releases. The timing is no coincidence, as companies compete to capture attention during the Spring Festival holiday period, when hundreds of millions of Chinese consumers are glued to their phones.

Video Generation Takes Centre Stage

ByteDance launched Seedance 2.0 on February 12, a next-generation video model built on a unified multimodal architecture that handles text, image, audio, and video inputs simultaneously. The model generates two-K resolution video with synchronised audio and multi-shot narrative coherence, drawing immediate comparisons to OpenAI's Sora 2. Hollywood has already pushed back, with Disney taking legal action over generated clips featuring copyrighted characters.

Kuaishou Technology unveiled Kling AI 3.0 on February 5, featuring native four-K resolution at sixty frames per second, multi-shot storyboarding, and audio generation across multiple languages. Since launching in twenty twenty-four, Kling AI has served over sixty million creators.

Language Models and Robotics

Zhipu AI released GLM-5, a seven hundred and forty-four billion parameter model trained entirely on Huawei Ascend chips with zero dependency on NVIDIA hardware. The model claims industry-leading scores in coding and agentic performance, pushing Zhipu's shares up nearly thirty percent.

Shanghai-based MiniMax launched M2.5, a model that achieves eighty point two percent on SWE-Bench Verified, edging out GPT-5.2, while costing roughly one-twentieth the price of Claude Opus 4.6.

Alibaba's DAMO Academy entered the robotics AI space with RynnBrain, an open-source foundation model that introduces spatiotemporal memory and physical spatial reasoning to enable robots to navigate real-world environments.

The Seven Hundred Million Dollar User War

The releases coincide with massive promotional campaigns. Alibaba's Qwen App launched a three billion yuan Spring Festival giveaway, Tencent's Yuanbao App is distributing one billion yuan in digital red envelopes, and Baidu's Ernie Bot offers five hundred million yuan in rewards, making this the most expensive AI user acquisition battle in history.

Published February 15, 2026 at 4:33am

More Recent Episodes