Meta has moved ahead of Nvidia in U.S. open-source AI model rankings, according to scores cited from the Artificial Analysis Intelligence Index. Nvidia’s Nemotron 3.5 Super VL briefly became the top U.S. open-source model with a score of 26.5, but Meta’s newly released Muse Spark Open soon overtook it at 39, a lead of 12.5 points.
The broader global open-source ranking still places several Chinese models ahead of Meta. Xiaomi’s MiMo-V2.6-Pro leads with 46 points, followed by Zhipu’s GLM-5.3 at 45. Muse Spark Open is tied with DeepSeek V4.1 Flash at 39. Meta’s new open model also sits close to its own closed-source flagship, Muse Spark 1.2, which scored 40 after launching in August.
The score breakdown points to a mixed picture. Among the nine highest-scoring open-source models, Muse Spark Open ranked last on a combined set of four agent tests, while its other six tests together placed fourth. The full 6-point gap versus GLM-5.3 came from the agent category. At the same time, Muse Spark Open led all open-source models in the AA-Omniscience knowledge accuracy test with 25 points, posted a 24% hallucination rate, and used fewer tokens per test, helping lower average testing cost to $0.84.
Nvidia’s hold on the top U.S. open-source model ranking lasted only a few hours, based on the Artificial Analysis Intelligence Index cited in the report.
Nemotron 3.5 Super VL scored 26.5, putting Nvidia ahead of every previous U.S. open-source model. Meta then released Muse Spark Open, which scored 39 and moved past Nvidia by 12.5 points to become the new U.S. leader.
Chinese models still lead the global open-source table
On the global open-source leaderboard, Meta still sits behind several Chinese models. Xiaomi’s MiMo-V2.6-Pro leads with a score of 46, while Zhipu’s GLM-5.3 follows at 45. Muse Spark Open is level with DeepSeek V4.1 Flash at 39.
The new open-source release is also close to Meta’s closed-source flagship, Muse Spark 1.2, which was launched in August and scored 40.
Agent tests were the main weak spot
Among the nine highest-scoring open-source models, Muse Spark Open ranked last on a combined set of four agent tests. Across the other six tests, it would rank fourth. Its 6-point gap with GLM-5.3 came entirely from the agent category.
Strength in knowledge accuracy and token efficiency
In the AA-Omniscience knowledge accuracy test, Muse Spark Open led all open-source models with 25 points. Its hallucination rate was 24%, the lowest among open-source models that scored above 30.
Token efficiency was also a relative strength. Muse Spark Open generated about 40,000 tokens on average per test, 43% fewer than GLM-5.3. The two models have similar API pricing, but Muse used fewer tokens, bringing its average cost per test to $0.84. That was less than half the cost of GLM-5.3, though still about three times the level of DeepSeek V4.1 Flash.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
