Chinese AI Models Cut US Performance Gap to 6% as Zhipu Tops Global Coding Rank
Updated
Updated · Bloomberg · Jul 26
Chinese AI Models Cut US Performance Gap to 6% as Zhipu Tops Global Coding Rank
3 articles · Updated · Bloomberg · Jul 26
Summary
June data showed Chinese large language models trailing US peers by a record-low 6%, narrowing from 9% in May and well below the roughly 10%-15% gap seen over the prior year.
LiveBench rankings drove that shift: Chinese models captured two global top slots last month, with Zhipu AI’s GLM-5.2 reaching No. 1 in agentic coding.
Bloomberg Intelligence said Moonshot’s gains suggest Zhipu was not a one-off, raising questions about how durable US technological supremacy in AI will be.
The latest rankings point to broader momentum in China’s AI sector, not just a single standout model, as competition with US developers tightens.
With foreign AI models slashing costs by 90 percent, could U.S. tech giants lose market dominance despite maintaining a slight performance edge?
If benchmark scores mask critical real-world failures, is the shrinking global AI gap a true technological triumph or just an illusion of progress?
From 6% to Disruption: How Chinese Open-Weight AI Models Redefined the US-China AI Race in 2026
Overview
In July 2026, the US-China AI race reached a turning point as Moonshot AI prepared to release the open weights of its massive Kimi K3 model, following a record-low 6% performance gap between the two countries. By publishing Kimi K3’s weights on Hugging Face, Moonshot enabled global developers to self-host or use its API, reducing reliance on US closed-source providers. This breakthrough, powered by innovations like sparse Mixture-of-Experts and efficient attention mechanisms, triggered both a surge in Chinese AI capabilities and intense US accusations of model distillation. The rapid progress, combined with US export controls and China’s shift toward algorithmic efficiency, sparked global market turmoil and deepened regulatory and compliance anxieties for Western enterprises.