Meta unveils Muse Spark 1.3 with coding improvements

Meta Platforms has officially launched Muse Spark 1.3, an upgraded artificial intelligence model aimed at delivering frontier-level performance at lower operational costs. According to CEO Mark Zuckerberg, the update marks Meta's most substantial advancement to date in software coding and autonomous agentic workflows. The model has been made available via Muse Code and Meta's API, with future plans to release open-weights versions. In industry benchmark evaluations, Muse Spark 1.3 demonstrated strong results across complex long-context and developer tasks. It recorded benchmark scores of 98.5 on MRCR 256K–512K and 98.1 on MRCR 512K–1M, leading its comparison class. Additionally, the model scored 75.4 on DeepSWE v1.1, 59.4 on SWEAtlas CodeBase QnA, and matched OpenAI's GPT-5.6 with an 88.8 score on Terminal-Bench 2.1. Following the product release, Meta's shares traded higher by 2.56% at $593.38. The launch underscores Meta's ongoing competitive push against frontier AI labs by enhancing its developer ecosystem and integrating specialized coding agents into commercial workflows.