The gap between the best closed frontier models from US companies and the leading open-weights models from China has shrunk to roughly four months, according to Mozilla's State of Open Source AI report. This rapid convergence means that paying for expensive frontier models now buys a relatively short performance head start. The report indicates that most organisations should default to open models for the majority of their workloads, reserving paid frontier models for specific, high-intensity tasks.

Mozilla's analysis highlights Moonshot AI's Kimi K3 as a standout open model. It achieves a performance score on the Artificial Intelligence Index that is just three points behind Anthropic's Fable 5, all while costing only 30 percent of the closed model. This stark difference in price versus performance suggests a clear economic incentive for companies to adopt open-source alternatives where possible.

Raffi Krikorian, Mozilla's chief technology officer, notes that the decision to pay for closed models is workload-specific rather than an organisation-wide strategy. He identifies three key areas where frontier models earn their premium: expert professional work, high-intensity retrieval, and long context handling. This distinction helps businesses optimise their AI spending by targeting expensive models only where they provide a tangible advantage.

The report underscores a strategic shift in the AI industry. As open models close the gap, the value proposition of frontier models is increasingly defined by niche capabilities rather than raw performance superiority. This trend challenges the traditional dominance of closed-source providers and forces companies to re-evaluate their AI infrastructure investments based on specific use cases rather than general capability.