Performance Gap Between US Closed-Source and Chinese Open-Weight AI Models Narrows

A report by Mozilla titled 'State of Open Source AI' indicates that the performance gap between US closed-source frontier models and Chinese open-weight models has narrowed to approximately 4.4 months. Models such as Moonshot AI's Kimi K3 are trailing Anthropic's Claude 3.5 Sonnet (referred to as Fable 5 in the report) by only three points on the Artificial Analysis Intelligence Index, while operating at roughly 30% of the cost. METR assessments show that while top closed-source models can complete 12-hour expert tasks, the best open-source models are currently capable of 7-hour tasks, with projections suggesting they will match current closed-source capabilities in about four months.