How far behind is open-source AI? (yaroslavvb.github.io)

A recent analysis has highlighted the evolving landscape of open-source AI models in comparison to proprietary counterparts, revealing a narrowing gap in performance. The study tracked the advancement of open-weight models relative to proprietary models, using the Artificial Analysis Intelligence Index (AI Index v4.1) as a consistent benchmark. The data indicates that the performance gap peaked at around 10 months with the release of DeepSeek V3 in December 2024, but has since decreased to approximately 2–3.5 months, with emerging models like DeepSeek, Kimi, Z AI, and MiniMax making significant strides. This trend is crucial for the AI/ML community as it signals a positive shift towards the competitiveness of open-source models in a domain historically dominated by proprietary technologies. With 23 models qualifying as frontier-pushers—defined as those whose AI Index surpassed all predecessors—the open-source movement is gaining momentum. This shift not only enhances accessibility to advanced AI capabilities but also spurs innovation, encouraging further development within the open-source community. As the capabilities of these models continue to improve, they may challenge proprietary systems at various benchmarks, potentially reshaping the future of AI development.