Open AI models catch up to frontier in 4 months

A report from Mozilla, previewed exclusively by Ars Technica, finds that open-weight AI models — many of them Chinese — now match the capabilities of top closed frontier models within about 4 months of their release. Access to the very latest flagship model, though, still costs roughly 5 times more than a comparable open alternative.

That reshapes the calculus for anyone buying AI: paying a premium no longer buys a fundamentally different quality tier, just a few months' head start before open models catch up. For companies building products and workflows on AI, the gap between the expensive and the free option has shrunk from a lasting edge to a single quarter's window.

Check which model your current AI tools actually run on, and whether that choice is driven by real speed and accuracy gains or just habit. If a task doesn't need the very latest model, shifting it to a cheaper option can free up budget for new use cases without hurting quality. Matching the right engine to the task is the same idea behind 1Cifer: agents assigned by role draw on the skills needed for each job instead of locking into one costly model.