Nvidia has just won the top spot in open source in the United States, but Meta has surpassed it with a score of 39, while Agent lags behind Chinese models in terms of score Dynamic Beating AI News: Nvidia's US open-source model ranked first and only lasted for a few hours. Nemotron 3.5 Super VL scored 26.5 on the Artificial Analysis Intelligence Index, surpassing all previous open source models in the United States. But Meta's subsequent Muse Spark Open scored 39 points, leading Nvidia by 12.5 points and becoming the new number one in the United States. However, looking at the global open source model rankings, Meta still lags behind several Chinese models. Xiaomi MiMo-V2.6-Pro leads with 46 points, while Zhipu GLM-5.3 has 45 points, and Muse Spark Open and DeepSeek V4.1 Flash have the same 39 points. It is close to Meta's closed source flagship Muse Spark 1.2, which was launched in August and scored 40 points. Among the top 9 open-source models with the highest scores, Muse Spark Open's 4 Agent tests performed last overall, while the other 6 tests combined ranked fourth. The 6 points that differ from the GLM-5.3 of Zhipu are all from Agent testing. On the contrary, it leads all open source models with a score of 25 in the knowledge accuracy test AA Omniscience, with an illusion rate of only 24%, and is the lowest among open source models that score over 30 points. Muse Spark Open's token usage efficiency is also good. On average, each test outputs about 40000 tokens, which is 43% less than GLM-5.3. The unit prices of the two APIs are similar, but due to Muse consuming fewer tokens, the average cost per test is only $0.84, less than half of GLM-5.3, but still about three times that of DeepSeek V4.1 Flash.