Meta Releases Muse Spark 1.3 with 75.4% DeepSWE Score

Meta Superintelligence Labs released Muse Spark 1.3, setting new efficiency records for autonomous software engineering agents. The model achieves a 75.4% score on DeepSWE 1.1 and 88.8% on Terminal-Bench 2.1, resolving complex codebase refactoring tasks using roughly 20% fewer tool invocations and 25% fewer total prompt tokens than rival foundation models.