Citation

Zainal Abidin, Mohamad Izani and Hadi Ghasemi, Mohammad and Abdul Razak, Norainy and Abdul Razak, Aishah (2026) Intelligence Without Borders: Empirical Evidence of US-China LLM Convergence and the Ecosystem-Dependent Licensing Gap. In: 2026 International Conference on Integrated Intelligence and Cognitive Engineering, ICIICE 2026, 18 April 2026 - 19 April 2026, dubai.

Text 14.pdf - Published Version Restricted to Repository staff only Download (3MB)

Abstract

The release of DeepSeek-R1 in January 2025 reframed global AI competition by demonstrating frontier-level intelligence from a Chinese ecosystem model at a fraction of the cost attributed to US alternatives. Despite extensive policy commentary, quantitative empirical evidence characterising the performance structure of US versus Chinese LLMs remains scarce. This paper investigates whether statistically significant capability differences persist between national ecosystem tiers across 188 large language model configurations (137 confirmed after intelligence-score exclusions) from 37 creators, benchmarked against the 2026 Artificial Analysis dataset. Using Pareto frontier analysis on the intelligence–cost plane, Kruskal–Wallis tests with Holm-corrected Dunn’s post-hoc comparisons, and stratified Mann–Whitney licensing decomposition, we find that US Frontier and Chinese Ecosystem models form a statistically indistinguishable upper intelligence tier (Dunn p = .206, Holm-corrected) while significantly outperforming Enterprise Cloud, Specialist Lab, and APAC/MENA tiers (η²H = .233). Chinese ecosystem models occupy 4 of 8 Pareto-optimal positions, delivering 80% of US frontier intelligence at under 7% of the cost. The aggregate commercial–open-source gap (d = 0.84) conceals an ecosystemdependent effect: large in US Frontier (d = 1.74), present in the Chinese Ecosystem (d = 1.15), and reversed in the Specialist Lab tier (d = −1.24). Findings carry direct implications for AI procurement strategy, multimedia system deployment, and the evolving geopolitics of technological capability.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	large language models, ecosystem benchmarking
Subjects:	H Social Sciences > HC Economic History and Conditions > HC94-1085 By region or country
Divisions:	Faculty of Business (FOB)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	04 Aug 2026 06:48
Last Modified:	04 Aug 2026 06:48
URII:	http://shdl.mmu.edu.my/id/eprint/16495

Downloads

Edit (login required)