Open AI models narrow the frontier gap to 4.4 months

Comparing capability gaps and cost per task

Mozilla’s latest State of Open Source AI assessment, as described in the article, says Moonshot AI’s Kimi K3 scored three points behind Anthropic’s Fable 5 on the Artificial Analysis Intelligence Index while costing 30 percent as much. The article also cites a neutral-harness Terminal-Bench 2.1 evaluation in which Z.ai’s GLM 5.2 scored within one point of Anthropic’s Claude Opus 4.7 and 4.8, at about one-fifth of the per-task cost. The reported remaining advantage for closed frontier models is concentrated in tasks taking roughly eight to 12 hours for human experts to complete.

Closed frontier models retained a capability lead over leading open-weight models, particularly for longer and more demanding tasks.

Mozilla’s assessment puts the remaining gap at 4.4 months: for a narrow band of tasks, paying for a closed frontier model reportedly buys roughly a four-month lead at about five times the per-task cost.

If the assessment holds up, model selection becomes less of an organization-wide choice between open and closed systems and more dependent on the workload. It also points to a small set of time-sensitive or demanding tasks where the premium for a closed model may still matter.