According to Perspicacity Beating monitoring, after Meta released Muse Spark 1.2, the evaluation agency Artificial Analysis gave it a score of 54, which is 3 points higher than the previous version and 11 points higher than the initial version in April this year. It is now tied with Grok 4.5 and is only one point behind GPT-5.5, placing it joint third among U.S. AI models.
This progress is mainly attributed to the Agent task. In the real-world knowledge work evaluation GDPval-AA v2, Muse Spark 1.2 has risen from 1371 Elo to 1631, jumping to fifth place overall, surpassing Claude Opus 4.8. There have also been improvements in end-user operations and bank tool invocation, while the performance in scientific reasoning has remained essentially unchanged.
There has been no price increase. The price remains at $1.25 and $4.25 per million input and output tokens, respectively. Artificial Analysis estimates that the average cost for completing a comprehensive evaluation task is $0.40. GPT-5.5 costs $1.18, and Kimi K3 costs $0.86.
