The release of DeepSeek’s V3 and R1 models with open weights could be a turning point in artificial intelligence (AI) development, representing meaningful progress in making it more accessible and transparent. However, the Open Source Initiative’s comprehensive definition of Open Source AI (OSAI) offers a compelling vision of what even greater transparency could achieve. Drawing from DeepSeek’s recent innovations in model optimisation and efficiency, we present three interconnected arguments supporting the evolution from current open-weight practices toward full OSAI development.

First, we examine how DeepSeek’s success with publicly available weights demonstrates the viability and value of increased transparency. We highlight the additional benefits of embracing the entire OSAI framework, which includes training data provenance and characteristics, complete source code for training and inference, and full model parameters. Second, we argue that pivoting toward genuine open-source standards can potentially de-escalate the emerging AI “arms race” by transforming a seemingly zero-sum game into a collective endeavour for advancement. Third, we contend that open-weight initiatives could naturally evolve toward the Open Source Initiative’s standards of transparency and accountability in AI systems – a critical consideration as we approach potentially superintelligent systems where value alignment becomes paramount. The point of convergence for the recommendations will be global cooperation to establish a framework for responsible AI governance – one that effectively balances innovation with accountability.

As AI systems increasingly train successive generations of models, this policy brief argues that building upon current open-weight practices toward comprehensive open-source standards is necessary for the responsible advancement of the technology – aligning with the G20’s AI governance and global safety standards. By advocating for stronger open-source AI practices, the G20 can play a crucial role in tech diplomacy, de-escalating the emerging AI arms race. With AI becoming increasingly significant in shaping economies and societies, fostering collaboration over proprietary competition also resonates with the G20’s broader vision of inclusive and sustainable technological development.