Curated reading list outlines state of open models and their implications

Nathan Lambert of Interconnects has published a curated reading list intended to help readers quickly grasp the state of open models, their strategic implications, and the ongoing debates around their risks and benefits. The list spans essays, research papers, and policy analyses published between 2023 and 2026, and is designed to serve as a comprehensive overview for those seeking to understand the open-model landscape.

The list includes perspectives on why organizations release open models, such as Meta’s rationale for open-weight releases like Llama 3, and frameworks for thinking about openness on a spectrum rather than as a binary choice. It also features analyses on the economic role of open models, their impact on innovation, and their potential to enable custom agentic workflows in enterprises.

Safety and risk are recurring themes, with entries discussing how open models compare to closed models in terms of safety guardrails and the marginal risks associated with open-weight releases. The list also highlights concerns about the decline of open data and the challenges of regulating open models, including the difficulty of preventing misuse by bad actors.

Technical evaluations are a core component of the list, including independent analyses of the performance gap between open and closed models. Recent assessments suggest the gap has narrowed to roughly four to six months, with leading open models increasingly originating from Chinese labs since 2024. The list cites SemiAnalysis’ independent evaluations and cost-performance analyses from Artificial Analysis to support these claims.

The reading list also documents the role of Chinese labs in advancing open models and their growing adoption by Western companies. Examples include Western firms like Perplexity and Thomson Reuters shifting from closed to open Chinese models such as DeepSeek R1 and Qwen, respectively. The list includes regulatory and media coverage of these shifts, as well as analyses of China’s structural advantages in open-source AI development.

The list was last updated on September 13, 2026, and Lambert invites readers to suggest additional pieces for inclusion. It also references tools and dashboards maintained by Interconnects, such as the Adoption Dashboard and Artifacts Hub, which track model downloads, derivatives, and research adoption by region.