Proprietary Frontier AI Offers Just Four-Month Lead Over Open Models at 5x Cost
A preview of an upcoming Mozilla report reveals that proprietary frontier AI models maintain only a four-month capability lead over open-weight alternatives. Despite this narrow advantage, using closed proprietary models costs roughly five times as much as running open-weight options. This narrowing performance gap undermines the economic rationale for paying high subscription or API fees for closed-source models. It signals a shift where enterprises can leverage increasingly capable open-weight AI—often led by open releases from Chinese research teams—at a fraction of the cost. The report highlights that open-weight alternatives quickly replicate the core reasoning and coding capabilities of leading frontier models. As a result, commercial vendors face growing pressure to innovate faster or lower their prices to justify the steep cost premium.
## BACKGROUND
Frontier AI models represent the most advanced general-purpose systems, typically trained by major tech labs using vast compute resources and accessible only via proprietary cloud APIs. Open-weight models release their final trained neural network parameters publicly, allowing developers to download, self-host, and fine-tune the models on their own hardware.