Mozilla's inaugural "State of Open Source AI" report, published on September 15, shows that open-weights models are no longer merely playing catch-up. The performance gap between open models and top proprietary systems has narrowed to just 3%, while the cost of using these models has fallen by up to 50x over the past three years.
Performance Gap Between Open-Weights and Proprietary AI Models Narrows to 3%, Mozilla Report Finds
The report highlights a growing economic disparity: while open models power roughly one-third of real-world AI usage, they currently account for only 4% of total AI revenue. This economic gap persists despite high adoption rates; for instance, eight of the top 10 models ranked by token volume in August 2026 were open-weights models.
Geopolitically, the report notes that China and East Asia lead the world in open-source AI adoption at 89%. Mozilla's Chief Technology Officer, Raffi Krikorian, observed that while open models offer infrastructure sovereignty, the current open ecosystem is largely supported by Chinese capital. Krikorian emphasized the need for diverse investment and infrastructure to ensure technological plurality and prevent a concentration of power in closed, proprietary systems.
Sources
- Exclusive: Paying for frontier AI models buys 4-month head start at 5x the cost （Ars Technica AI, 2026-09-15）
- Mozilla