Meta's Muse Spark 1.3 cuts coding token use by 25% and boosts agentic reliability
Introducing Muse Spark 1.3
Meta releases Muse Spark 1.3, an AI model update for coding and agentic tasks. It improves long-horizon task execution, better follows complex instructions, and reduces tool calls by ~20% and tokens by ~25% in coding workflows. The model also shows stronger resistance to adversarial inputs and better calibration for irreversible actions. Available in Muse Code and Meta Model API.
When given an open-ended objective, it uses tools to generate its own context across messy and conflicting sources, proactively corrects gaps in its plan, and keeps track of what it has learned to produce a final deliverable.