The Sequence Opinion #762: Trillion-Parameter Diplomacy: China, the US, and the Battle for Open Models
A practical overview about the state of US-China open source AI race.
Happy Thanksgiving Holiday for those in the US. I have many things to be thankful for this year including this incredible audience. Today, we are going to debate a hot topic: US vs. China in open source AI.
The past two years have seen a seismic shift in open-source AI, with both Chinese and Western organizations racing to release ever more capable large language models (LLMs). What began with a few leaked research models has exploded into a global competition to push the boundaries of what open models can do.
In this essay, we’ll dive into the state of this race as of late 2024–2025, comparing the leading contributions from China and the US (broadly including Western efforts). We’ll explore major model families on each side – from Alibaba’s Qwen/Tongyi and Moonshot’s Kimi in China, to Meta’s LLaMA series and OpenAI’s GPT-OSS in the West – and analyze how they stack up in architecture, capabilities, openness, performance, community adoption, and more. We’ll also dedicate a section to a strategic question: can the US open ecosystem actually pull ahead from here, or has the center of gravity already tilted toward China?
Think of this as a walkthrough of the open LLM landscape: technical, opinionated, and zooming in and out between micro details and the big picture.
For a deeper analysis on this topic, check out the US vs. China open source AI evaluation dashboard released by my company LayerLens: https://app.layerlens.ai/evaluation-space/6927234d10bd104696db7f44