Musk Intensively Likes China's AI: Kimi, ByteDance, and Qwen All Gain Top-Tier Recognition

21st Century Business Herald Reporter Lei Chen

Recently, Tesla and xAI founder Elon Musk have been focusing on China’s AI sector, repeatedly praising domestic models on social media and public occasions. Among them, Kimi, ByteDance Seedance 2.0, and Alibaba Qwen 3.5 have all been listed and received official responses, creating a transoceanic tech interaction.

On March 16, Kimi released the technical report “Attention Residuals,” which reconstructs the residual connection mechanism of large models. Using the new “attention residual” method, it achieves simultaneous improvements in training efficiency and core performance—training efficiency on a 48B parameter model increased by 1.25 times, with scientific reasoning and mathematical scores improving by 7.5% and 3.6%, respectively. This has been regarded by the industry as an important signal of “Deep Learning 2.0.”

In response, Musk immediately reposted and commented on X, “Impressive work from Kimi.” Kimi’s official account humorously replied, “Your rockets are pretty good too!”

The reporter notes that prior to this, Alibaba Qwen 3.5 series models gained Musk’s favor with their “extreme intelligence density.” On March 2, Alibaba Qianwen officially open-sourced four smaller models: Qwen 3.5-0.8B, 2B, 4B, and 9B, covering all scenarios from edge devices to lightweight servers. These models achieve advanced performance with minimal parameters, breaking the stereotype of “small models with weak capabilities.” The 9B version performs comparably to models with hundreds of billions of parameters, while the 0.8B and 2B versions can run smoothly on mobile and IoT edge devices. Musk directly commented under the official Qwen X post, “Impressive intelligence density.”

Additionally, ByteDance’s next-generation video generation large model Seedance 2.0 began internal testing on February 12. With a unified multimodal audio-visual joint generation architecture supporting text, image, audio, and video inputs, it addresses industry pain points such as low usability of AI video generation and character detail drift. It can generate up to 60 seconds of 2K broadcast-quality videos. Musk later reposted the related tweet and remarked, “It’s happening fast.”

Beyond praising specific models, Musk also clearly predicted at the Davos Forum and in podcast interviews that China’s AI computing power will far surpass other regions worldwide. Stable and inexpensive electricity, large-scale infrastructure, and an efficient team of engineers are China’s core advantages in AI.

Industry analysts believe that Musk’s continuous praise for domestic models demonstrates China’s comprehensive breakthroughs in AI—from underlying architecture and multimodal applications to open-source ecosystems.