AI Labs & Models

A16Z partners argue open source AI models outpace closed systems

A16Z partners claim open source models are lapping closed systems in capability and adoption. They argue this shift is essential for safety research and preventing the centralization of AI power.

· 1 source · Open source momentum

A16Z partners argue that open source AI models are currently lapping closed models by six to 12 months in capability. Mark Andreessen stated that the current U.S. administration is 100% pro-open source and wants American open source to win globally. He noted that the Chinese Communist Party is also actively supporting open source AI, contrasting this with the EU, which he said still aspires to regulate or ban open source but lacks leverage. Andreessen emphasized that open source is essential for safety, allowing every PhD student and professor to inspect model weights to solve issues like reward hacking, which labs are currently neglecting. He also highlighted the educational value, arguing that without open source, understanding AI would be restricted to employees of a few large companies.

Chris Dixon noted that open source models are growing faster in token consumption than proprietary ones, as seen in OpenRouter data. The partners view open source as the guarantor against the centralization of AI power in one or two companies or the government. They consider it critical for safety research and public education. Andreessen described the administration's stance as very encouraging, stating it is 100% pro-open source.