Meta released Muse Spark 1.3, its latest model for agent and coding work. It is available today in Muse Code and the Meta Model API. Earlier reasoning modes ship now. A new max reasoning mode follows once safety testing finishes. Meta says the model sustains longer tasks and tracks work across a single long thread. It asks clarifying questions, requests help when stuck, and confirms before consequential actions. Meta engineers measured fewer tool calls and fewer tokens than version 1.2 on coding tasks. Meta also reports stronger resistance to prompt injection and better judgement on irreversible actions. Meta says bigger models and an open weights release of Muse Spark are planned.
What changed
Muse Spark 1.2 used more turns, more tokens and was more verbose in coding work.
What it unlocks
Running longer multi-step agent tasks that ask clarifying questions before acting.
- ~20% fewer tool calls
- ~25% fewer tokens
What you need to act on it
- access to Muse Code or the Meta Model API
Sources