The State of Open-Source AI and Why It Matters

Logistical Note: This was a fully in-person event. Lunch was available for attendees.

Open-source AI models have sharply closed the performance gap with closed systems this year, with top open-weight models narrowing the difference from about 8 percent in early 2024 to 1.7 percent by February 2025. Because open models can be run and modified directly, rather than only accessed through an API, they enable a broad set of research and development activities, including testing how models perform on different computing hardware, adapting them to specific tasks, and examining how they behave internally.

The in-person panel held on Capitol Hill explored the unique use cases enabled by open models, how open-source AI increases competition across the AI stack, and how it enables a wide range of firms, researchers, and public institutions to build and deploy AI. It also explored what steps policymakers can take to support a secure and thriving open AI ecosystem.