Meta releases Muse Spark 1.3 with 20% fewer tool calls and 25% fewer tokens

Meta Superintelligence Labs released Muse Spark 1.3 on September 2, the fourth Muse Spark release in five months. The model targets long-horizon agentic and coding work and is available the same day in the Muse Code terminal agent and the Meta Model API under the model ID muse-spark-1.3.

On coding efficiency, Meta engineers reported approximately 20% fewer tool calls and 25% fewer tokens compared to Muse Spark 1.2 for equivalent tasks. On DeepSWE v1.1 long-horizon software engineering, Muse Spark 1.3 scored 75.4, ahead of Claude Opus 5 at 74.0 and GPT-5.6 Sol at 73.0. Terminal-Bench 2.1 reached 88.8, tying GPT-5.6 Sol, and long-context MRCR at the 512K-to-1M band hit 98.1.

On collaboration, the model now asks clarifying questions on ambiguous prompts, calls the user in when stuck, and confirms before taking consequential actions, with improved resistance to prompt injection. Standard API pricing remains unchanged at $1.25 input and $4.25 output per 1M tokens. A Contributor tier drops to $0.10 input and $0.20 output in exchange for letting Meta train on traffic. The model remains closed-weight with no open-weights date announced.