The performance gap between frontier AI models from US tech companies and the best open-weights models from Chinese companies has closed to just 4.4 months, according to a new Mozilla report previewed by Ars Technica. That narrow margin helps explain why many organizations are shifting to significantly cheaper open models for routine work.
The report reveals a striking cost disparity: paying for frontier model access buys you roughly four months of performance leadership, but at five times the cost of the best open alternatives. For most organizations, the math increasingly favors open models for the majority of workloads, reserving expensive frontier access only for tasks where that marginal improvement genuinely matters.
Mozilla's analysis identifies a narrow band of workloads where frontier models remain worth the premium — primarily complex reasoning tasks, novel edge cases, and safety-critical applications where the extra capability justifies the cost. For everything else, open models have become the rational default.
The findings have significant implications for the AI business model. If open models continue closing the gap at this pace, the window for recouping massive R&D investments through API pricing shrinks further. Companies like OpenAI, Anthropic, and Google face growing pressure to either dramatically accelerate their research pace or find other ways to differentiate beyond raw model performance.
For the enterprise market, the report is a practical guide. Most organizations should be running open models as their default workhorse and reaching for frontier APIs only when the task specifically demands it.