The performance gap between frontier AI models built by US tech companies and the best open-weights models from Chinese companies has narrowed to just 4.4 months. That is the central finding of a new Mozilla report, and it explains why a growing number of companies are switching to open models for everyday tasks.
The math is simple. Frontier models cost about five times more than open alternatives, but they only buy you roughly a four-month head start in capability. For most routine work, that trade-off no longer makes sense.
Mozilla Report: Open Models Close the Gap to 4.4 Months
According to the latest State of Open Source AI report from Mozilla, most organizations should be using open models as the default for the majority of their work. The report was published on September 15 and shared with Ars Technica prior to publication.
The report points to Moonshot AI's Kimi K3, a leading open model, which achieves a composite AI performance score on the Artificial Analysis Intelligence Index that is just three points below frontier models. That is a remarkably small gap given the cost difference.
"Most organizations should ideally be using open models as the default for the majority of their work." — Mozilla, State of Open Source AI report
Why Companies Are Choosing Open Models Over Frontier AI
The cost factor is driving the shift. When an open model performs nearly as well as a frontier model but costs a fraction of the price, the business case for paying premium rates weakens considerably.
- Frontier models cost approximately 5x more than open-weights alternatives
- The performance gap has shrunk to 4.4 months
- Kimi K3 scores just three points below frontier models on the Artificial Analysis Intelligence Index
- Mozilla recommends open models as the default choice for most organizational workloads
The report does not suggest frontier models are useless. Instead, it reveals a narrow band of workloads where paying the premium is still worth it. For everything else, open models deliver comparable results at a much lower cost.
Our Take: The AI Cost Equation Has Fundamentally Changed
To put it plainly, this report should make every technology decision-maker pause and reconsider their AI spending. If you are paying five times more for a four-month capability lead, you need to ask yourself a hard question: does that four-month advantage actually matter for the work your team does every day?
In our view, the answer is no for most companies. The era of defaulting to the most expensive frontier model is ending. Mozilla's data makes it clear that open models have caught up enough to handle the bulk of real-world tasks. The smart move is to use open models as your workhorse and save frontier models for the specific, high-stakes problems where that extra edge genuinely counts.
Companies that fail to make this shift are not just overspending — they are misallocating resources that could go toward other priorities. The 4.4-month gap is a number every CTO should have on their whiteboard right now.
