The performance gap between closed source frontier models in the United States and open source weighted models in China has narrowed to about 4.4 months

The "State of Open Source AI" report released by Mozilla shows that the performance gap between the US closed source frontier model and the Chinese open source weight model has narrowed to about 4.4 months. Moonshot AI's Kimi K3 only lags behind Anthropic Fable 5 by three points in the comprehensive score of Artificial Analysis Intelligence Index, with a cost of about 30% of Anthropic Fable 5. METR calculations show that the strongest closed source model can stably complete human expert tasks with a duration of about 12 hours, while the strongest open source model can complete tasks with a duration of about 7 hours. It is expected that the open source model will catch up with the current closed source level in about 4 months. (Source: Ars Technica)