Meta Unveils Muse Spark 1.1 for Multi-Agent AI

Meta has launched Muse Spark 1.1, a flagship large language model designed for multi-agent automation workflows. The model features a context compaction mechanism to preserve critical data across complex tasks and a 1 million token context window. It is available through Meta AI and the new Meta Model API in public preview. Muse Spark 1.1 scored 72.2 on the Vibe Code Bench v1.1 benchmark, over 50 points ahead of Meta's previous flagship model. Beyond coding, it supports multistep tasks like generating e-commerce listings and placing restaurant orders. The launch coincides with reports that Meta plans to expand data center capacity to 14 gigawatts using its custom Iris AI chip.