Funding policies can regulate the expenditure of article processing charges (APC) for gold open access (gold OA) papers, but cannot coordinate the management of different publishing models. Publishing models can, however, be compared systematically along the dimension of paper impact. Based on the data of Chinese papers included in the Web of Science (WoS) database from 2015 to 2024, this paper adopts Spearman rank correlation and non-parametric Kruskal-Wallis test methods to explore the influence mechanism of publishing model evolution on international collaborative papers. Chinese SCI papers are categorized into four groups based on the two-dimensional framework of “collaboration type × publishing model”, namely international collaboration with gold OA, international collaboration with non-OA, domestic collaboration with gold OA, and domestic collaboration with non-OA. A three-year rolling-average smoothing procedure is applied to improve the robustness of the correlation analysis. The empirical results reveal significant differentiated correlation characteristics between collaboration types and the academic influence of Chinese SCI papers. To ease the tension between these factors and paper impact, it is essential to strengthen international collaboration and prioritize the publication of high-potential highly-cited papers in Q1 journals; improving the content quality of domestically collaborative papers can systematically optimize core academic indicators, while citation improvement strategies for non-collaborative papers from single domestic institutions require strict effectiveness verification. Further analysis shows that structural adjustments in international publishing in recent years have raised the acceptance threshold for non-OA papers, leading to the concentration of China’s high-quality international collaborative resources in the non-OA publishing model, while the gold OA publishing model only serves as a fallback channel for suboptimal international collaborative resources in China. Such adjustments have also driven the expansion of non-collaborative, non-OA papers in China, and have particularly intensified the outflow of domestically collaborative non-OA papers — which deliver higher academic influence than domestic gold OA papers — and restrained the growth of internationally collaborative non-OA papers with high category normalized citation impact (CNCI). Such structural changes have triggered the one-way closed outflow of high-level academic papers through non-OA channels. This not only endangers national high-level scientific and technological information security, but also weakens the internal circulation efficiency of China’s OA knowledge innovation, further exposing the structural dilemma of “high output with insufficient high-quality achievement” of Chinese SCI papers. Therefore’s indigenous gold OA publishing system urgently needs to be deeply embedded in the global open science ecosystem so as to enhance the systemic capacity of China’s high-level opening-up. Finally, phased paper backflow strategies are proposed from the perspective of overall management of publishing models: constructing a domestic free diamond OA platform to attract the backflow of domestic collaborative non-OA papers and strengthen the internal circulation efficiency of China’s OA knowledge innovation; strengthening public oversight of the promotional performance of international platforms so as to build a sound competition mechanism among international publishers in terms of academic impact; and hedging backflow risks through an academic influence compensation mechanism.