Mozilla’s Inaugural ‘State of Open Source AI’ Report Is Here

Mozilla’s inaugural State of Open Source AI report, based on a survey of over 950 developers, concludes that open models no longer lag significantly behind proprietary systems like ChatGPT and Claude, with performance gaps narrowing to 3% and costs dropping up to 50-fold in three years. The analysis highlights that while open models now drive roughly one-third of AI deployments, they account for only 4% of industry revenue, indicating a disconnect between usage and financial returns within the open ecosystem.

The report reveals a stark regional divide in open source AI adoption, with China and East Asia leading at 89%, compared to lower rates in the West, where open source is increasingly treated as a national strategic asset. Governments worldwide are responding, with 12 new national AI strategies launched last year and 47 countries now restricting foreign processing for critical workloads, while the EU, Canada, and India are investing in open AI infrastructure as a public priority.

Despite 79% of developers using open models, only 51% have deployed them in production, compared to 63% for closed models, suggesting that infrastructure and tooling gaps, rather than model quality, are the primary barriers. Developers prioritize licensing terms (31%) and ownership (26%) when selecting models, reflecting a growing demand for control and flexibility over raw performance.

The report identifies the agentic harness—the software layer between users and models—as the critical factor shaping AI behavior, with minimal guardrails in place as users approve agent requests by default up to 93% of the time. Mozilla’s CTO Raffi Krikorian warns that without investment in open model infrastructure, governance, and tooling, restrictive closed systems may dominate, undermining public interest and tech sovereignty.