Open-Source AI & Open Models Reading List
| Source: Interconnects (Nathan Lambert)
Tags: open-source AI, open weights, Nathan Lambert, Kimi K3, GLM-5.2, Meta Llama, AI safety, open models
Nathan Lambert's updated reading list compiles the definitive works on open-source AI—covering open model strategy, release gradients, safety trade-offs, and China-US ecosystem dynamics—into a structured onboarding resource for practitioners entering the open weights space.
Details
Nathan Lambert (Interconnects) has updated his curated reading list on open-source and open-weight AI models, now featuring over 30 resources spanning foundational strategy, safety considerations, and recent developments through mid-2026. The list is organized thematically: why organizations release open models, how to think about release gradients (not binary open/closed), safety risk assessments, data commons erosion, and policy perspectives. Key 2026 inclusions reflect accelerating dynamics: Kimi K3 as an 'open-weights escalation' signal, GLM-5.2 as a step change for open agents, and Thinking Machines Lab's 'A Safe Path to Open Weights.' Lambert also cites his own analysis arguing open models are in 'perpetual catch-up' to closed frontier labs—a useful framing for enterprise procurement decisions. A notable thread through the list is the AI data commons decline (Longpre et al., 2024), which Lambert identifies as a structural constraint on truly open AI research, distinct from model weight availability. The list also covers Zuckerberg's articulation of Meta's open model strategy and Solaiman's gradient framework for evaluating openness. The reading list is a living document, not breaking news, but its breadth and curation quality make it the clearest single-stop reference currently available for understanding open model ecosystem dynamics. Particularly valuable for policy teams, enterprise architects evaluating vendor lock-in, and researchers entering the field.