The gap between open weights LLMs and closed source LLMs

	The gap between open weights LLMs and closed source LLMs(blog.doubleword.ai)
	234 points by kkm 15 hours ago \| 187 comments
	tl;dr: Analyzing the Artificial Analysis Intelligence Index suggests open-weight LLMs are closing the gap with closed-source models and could catch up by December 2026. However, expanding the analysis to all 18 benchmarks shows the average gap has stayed flat at roughly 5 months, with most of the headline improvement driven by coding benchmarks while other categories show slightly widening gaps.
	HN Discussion: •Open weights models are vulnerable as they depend on philanthropy from private orgs that can stop anytime ↑Chinese/open models depend on distillation from frontier closed models, creating a structural gap that won't close ↓Benchmarks are misleading because closed models can use backend augmentation systems beyond just weights •US export restrictions and closed approach may ironically be ceding ground to Chinese open-weight models ↓Article conflates open source with open weights, and these models are mostly Chinese rather than truly open