China AI Monthly: Model Gap Could Widen; Geopolitics -ve; Raised JEFc Capex

研报英文原文证据摘录

China AI Monthly: Model Gap Could Widen; Geopolitics -ve; Raised JEFc Capex

China (PRC) | Technology EquityJulyResearch13, 2026

Exhibit 1 - US-CN Model Intelligence GapChina AI Monthly: Model Gap Could Widen;

120%

Current gap: 15%

100%Geopolitics -ve; Raised JEFe Capex 80%

60% GLM 5.2 (85%) vs. Claude Fable 5 (100%)

40%In June, CN frontier models continued to narrow the performance gap

20%

0%to 15% and offer superior value-for-performance, but US labs' access to Dec-25 Jan-26 Feb-26 Mar-26 Apr-26 May-26 Jun-26 Jul-26

Source: Artificial Analysis, Jefferiesmore advanced GPUs and anti-distillation feature could widen the gap in .

2H26. Grok4.5 poses strong competition to top US/CN models sitting on Exhibit 2 - Avg Model Price - US vs CN

120%the Pareto frontier, while tighter geopolitics may slow model development 100% 100% 100% 100% 100% 100% 100% 100% 100%

& monetization. We also raised our 2026E-30E cuml China AI capex by 23% 100%80%

60%to US$1tn and power demand by 31%.

40% 29% 30% 31%

23% 18% 20% 22% 14% 14% 20%

US-CN model performance gap may widen in 2H26. In June, Fable 5 (Anthropic), GPT 5.5 (OAI) 0%

and GLM5.2 (Zhipu) were the top 3 globally on Artificial Analysis's (AA) intelligence ranking. Nov-25 Dec-25 Jan-26US AverageFeb-26price Mar-26 CNApr-26Average May-26price Jun-26 Jul-26

This is a milestone for China AI as it is the first time a Chinese model (also open source) entered .Source: Artificial Analysis, Companies, Jefferies

top 3 globally. But it did not last long as SpaceX released Grok4.5 on Jul 9 and overtook GLM5.2

as No. 3. Despite Fable 5's stellar capabilities, the US-China performance gap is still narrowing, Exhibit 3 - LLM's Intelligence to Price Ratio -

China Leads By Far

with GLM5.2 being 15% < Fable 5 (vs. 17% in May on consistent criteria).