Mozilla Report: Open-Weight AI Closes Capability Gap but Struggles with Revenue
A new assessment reveals that while open models now power a third of AI usage, they capture only 4% of industry revenue.
Mozilla released its State of Open Source AI 2026 (Volume 1.1) report in September, revealing a narrowing technical divide between open-weight models and closed frontier systems. The findings suggest that while open AI is becoming technically viable for mass adoption, a severe commercial imbalance persists.
According to the report, the capability gap between open-weight models and closed systems shrank from 8.04% to 3.3% over a 24-month period. This technical convergence is paired with a dramatic collapse in operational costs; inference for GPT-4-class models dropped approximately 50x over 36 months. Despite these gains, the financial rewards remain concentrated. Mozilla found that open models account for roughly 33% of real-world AI usage but capture only about 4% of the total industry revenue.
The Sovereignty Choice
This assessment arrives as open-weight models from developers like Mistral, DeepSeek, and Qwen challenge the dominance of closed-API providers such as OpenAI and Anthropic. Mozilla frames the shift toward open weights as a "sovereignty choice," allowing nations and enterprises to maintain control over their infrastructure and avoid vendor lock-in. Mozilla CTO Raffi Krikorian emphasized the importance of ownership in this transition, stating, "None of them asked permission, and none of them could have rented this. They own it. That is the whole idea."
Barriers to Production
Despite the narrowing capability gap, the report highlights a persistent struggle to move open models into live environments. Production readiness varies significantly by ecosystem: only 51% of teams using open models successfully reach production, compared to 63% of teams utilizing closed models. This suggests that the primary hurdles for open-source AI are no longer raw intelligence, but rather the operational complexities of security, compliance, and infrastructure management.
The Sustainability Risk
While open models have gained significant traction—now representing a majority of production tokens routed through the OpenRouter marketplace—the "money problem" poses a long-term risk. The decoupling of technical utility from commercial value means that the entities developing open-weight systems are not capturing the value they create for the end-user. If usage continues to scale without a corresponding increase in revenue share, the long-term sustainability of open-weight development could be threatened.
What to Watch
Industry observers will now be watching whether the 3.3% capability gap continues to shrink or stabilizes. The critical question remains whether the open-source community can develop a sustainable economic model that matches its technical success, or if the operational advantages of closed systems will keep the majority of high-value enterprise production locked behind proprietary APIs.