Introducing Muse Spark 1.3 (research.meta.ai)

Meta has rolled out Muse Spark 1.3, an upgraded model designed to enhance performance in agentic and coding tasks. This update is particularly significant for the AI/ML community as it advances the goal of achieving personal superintelligence, making the model more practical for real-world applications. Muse Spark 1.3 features improved multitasking capabilities and better communication with users, enabling it to manage longer-horizon workflows and navigate complex, conflicting information. It can now ask clarifying questions, adapt to user preferences, and follow detailed instructions more reliably, significantly reducing the risk of drifting from task requirements. In terms of coding, Muse Spark 1.3 shows marked improvements over its predecessor, executing coding tasks with increased efficiency—utilizing approximately 20% fewer tool calls and 25% fewer tokens. Additionally, the model exhibits enhanced adversarial robustness, better discretion on irreversible actions, and refined usability in engineering workflows. With these updates, Muse Spark 1.3 aims to push the boundaries of agentic AI capabilities, making it not only smarter but also safer for users. The model is now available through Muse Code and the Meta Model API, with further advancements on the horizon, including larger models and the release of open weights.