Meta's Muse Spark Open is now the strongest American-made open-weight model by a significant margin. Without it and Reflection's Beam, over 20 open-weight models built by Chinese companies would sit above the best Western alternatives. Mistral Large 4 has also been announced with an index score of 38, but its weights are not yet public.
The gap between American and Chinese open-weight models has been cut roughly in half this week. That is not a recovery. It is a shift from irrelevance to relevance, and the distinction matters. Chinese labs Kimi and Zhipu AI have been quiet since GLM-5.3 dropped August 18 and Kimi K3 dropped July 16, and nobody knows why.
The piece is worth reading in full for the competitive dynamics it maps: Meta dropped weights of a stronger model before Reflection and Mistral could release their own, the original article's analysis of why Western open models consistently attract more adoption than Chinese models at equivalent performance, and what Muse Spark Open actually means for the agentic ecosystem now that open-ended LLM agents are becoming real.
[READ ORIGINAL →]