Curated Reading List for Open‑Source AI and Open Models

Understanding the Rise of Open Models

Open‑source artificial intelligence and openly released model weights have become a focal point for researchers, policymakers, and businesses.

Nathan Lambert compiled a reading list on September 13 2026 to help anyone quickly grasp the state of open models.

The collection gathers essays, reports, and academic papers produced over the last few years.

Each entry explains a different facet of why organizations release models, how they fit into strategy, and what risks they pose.

One foundational piece is Bill Gurley’s “From Open Source Software to Open Source Strategy” published in May 2026.

Gurley walks through how open‑source software has historically been leveraged by companies and draws early implications for AI.

Mark Zuckerberg’s July 2024 commentary, titled “Open Source AI is the Path Forward,” offers a clear articulation of Meta’s rationale for releasing Llama 3.

He emphasizes that openness can accelerate innovation while expanding the ecosystem of developers.

Irene Solaiman’s February 2023 article, “The Gradient of Generative AI Release: Methods and Considerations,” argues that openness should be viewed on a spectrum rather than a binary closed/open divide.

She highlights factors such as licensing terms, compute cost, and data accessibility that shape each model’s openness.

Economic and Technical Implications

Nathan Lambert’s March 2026 essay “What comes next with open models” predicts that open models will complement strong closed models in future enterprise workflows.

He suggests that businesses will use open weights to build custom agentic pipelines tailored to specific tasks.

Christian Catalini’s August 2026 paper “Some Simple Economics of Open versus Closed AI” frames open models as a tool that can capture value across a broad swath of the economy, drawing parallels to historical intellectual‑property debates.

Lambert’s February 2026 analysis “Open models in perpetual catch‑up” warns that open models will likely lag behind closed models in raw performance.

He attributes the gap to the rapid scaling advantages enjoyed by well‑funded closed‑source teams.

In June 2026, Lambert explored divergent adoption curves in “Open and closed models are on different exponentials,” noting that open models may see slower uptake but can grow on separate trajectories.

Thinking Machines Lab’s July 2026 report “A Safe Path to Open Weights” proposes a balanced approach that releases powerful weights while embedding safety considerations.

Sayash Kapoor, Rishi Bommasani and co‑authors’ February 2024 study “On the Societal Impact of Open Foundation Models” identifies only marginal increases in documented risks for text‑focused large language models.

Florian Brand’s June 2026 article “The Myth of unsafe Open Source AI” counters that closed‑model safety guardrails are routinely bypassed, creating real AI‑risks before hypothetical open‑weight dangers materialize.

Shayne Longpre et al.’s July 2024 paper “Consent in Crisis: The Rapid Decline of the AI Data Commons” highlights a mass reduction in openly available training data, a key barrier to truly open AI research.

Geopolitical Landscape and Adoption Data

Nathan Lambert’s July 2026 piece “Kimi K3: The open‑weights escalation” examines how strong Chinese models influence the global AI ecosystem.

His June 2026 analysis of “GLM‑5.2 is the step change for open agents” underscores a significant capability jump among open‑weight Chinese agents.

The Golden Gate Institute for AI’s November 2025 presentation “Nathan Lambert on China’s AI Ecosystem and the Open Model Gap” offers a narrative of the 2025 open‑model story.

Quantitative snapshots appear in the April 2026 ATOM Report, which compares U.S. and Chinese open‑model adoption rates.

Interconnects’ own Adoption Dashboard tracks model downloads, derivatives, and research usage by region, providing the latest empirical evidence.

Finally, the Interconnects Artifacts Hub lists the most important models to know about in the current ecosystem.

Collectively, these resources give readers a comprehensive view of open‑source AI’s technical, economic, and policy dimensions.

Lambert invites readers to comment with additional pieces, promising to keep the list updated as the field evolves.

Why This Matters: The curated list equips stakeholders with the evidence needed to navigate open‑model opportunities and risks as they reshape the AI landscape.