← Latest AI News

August 10, 2026 · Daily brief

Open AI closes gap to 3%—but takes just 4% of revenue

Sovereignty angle
Performance parity means nothing if the revenue still flows to closed platforms. Open models run a third of production AI—but capture 4% of the money. The gap isn't technical anymore. It's who owns the deployment layer, the tooling, and the customer relationship.

Mozilla's first State of Open Source AI report shows the performance gap with proprietary models narrowed to 3%, while open models power a third of usage but earn 4% of revenue.

On July 14, Mozilla published its inaugural State of Open Source AI report, revealing that open models now trail proprietary systems like ChatGPT and Claude by just 3% in performance—down from an 8% gap in early 2025. The analysis, based on a survey of 950+ developers and 100 trillion tokens processed through OpenRouter, shows open models running one-third of real-world AI usage. Costs have dropped 50-fold in three years, with GPT-4-class inference now at $0.40 per million tokens versus $20 in 2023.

The revenue paradox

But here's the tension: open models capture only 4% of global AI revenue. Mozilla CTO Raffi Krikorian warns the battle has shifted from model weights to what the report calls the "agentic harness"—the deployment infrastructure, tooling, and governance layers where proprietary platforms still dominate. Half of developers use both open and closed models, but only 53% of open-model projects reach production versus 63% for closed, largely due to operational tooling gaps.

Why it matters

China leads open-source adoption at 89%, with models like Alibaba's Qwen hitting 942 million Hugging Face downloads—double Meta's Llama. Mozilla argues this is deliberate industrial policy: releasing weights globally while keeping inference local reduces dependency on U.S. platforms. For Latin American teams, the message is stark: technical capability is no longer the constraint. Control of the stack around the model is.

Sources