The performance lag between frontier AI models from US tech companies and the best open-weights models from Chinese companies has narrowed to just 4.4 months, according to a new report from Mozilla. This shrinking gap is driving many organizations to shift routine workloads to significantly cheaper open models, reserving expensive proprietary systems for specific high-value tasks.
What Happened
The latest State of Open Source AI report, published by Mozilla on September 15 and shared with Ars Technica prior to its public release, highlights that leading open models are now nearly on par with closed frontier systems. Specifically, the report cites Moonshot AI’s Kimi K3, which achieves a composite AI performance score on the Artificial Analysis Intelligence Index that is just three points behind Anthropic’s Fable 5 closed frontier model. Despite this near-equivalence in performance, Kimi K3 costs only 30 percent of what Fable 5 costs.
Why It Matters
The narrowing gap suggests that most organizations should ideally use open models as the default for the majority of their work, according to the Mozilla report. Raffi Krikorian, chief technology officer at Mozilla, explained that closed models earn their premium only in specific areas: expert professional work, high-intensity retrieval, and long context. "We see the decision to pay for closed as workload-specific rather than organization-specific," Krikorian said in an email to Ars. This distinction helps reveal a narrow band of workloads where frontier models remain worth the higher cost, while the bulk of routine tasks can be handled more economically by open alternatives.
The Bottom Line
With open Chinese models like Kimi K3 closing the performance gap to within three index points of US frontier models at a fraction of the price, enterprises are increasingly adopting a hybrid strategy. Routine operations are moving to cheaper open weights, while closed frontier models are reserved for specialized, high-intensity tasks.