Meta's Muse Spark 1.1 has improved its benchmark position and now edges past GLM-5.2 in coding, according to The Decoder. The report says the model reached a coding score of 71.3 while costing slightly less per task.
Coding has become one of the most competitive areas for model providers because it has measurable workflows and direct productivity value. Small gains in capability or cost can influence which models developers route into agents and IDE tools.
The update suggests Meta is pushing Muse Spark as a serious coding option, not just a general-purpose model family.