Muse Spark 1.3: 20% fewer tool calls, 25% fewer tokens

Meta released Muse Spark 1.3 on September 2, 2026, available with max reasoning on Muse Code and the Meta Model API.

The number that travels: in comparisons by Meta engineers against Muse Spark 1.2, it used about 20% fewer tool calls and about 25% fewer tokens, taking fewer turns where they were not needed and producing less verbose output.

The rest of the release is about behaviour in long-running work, which is the part that decides whether an agent is usable unattended:

• Sustains longer-horizon work in a single long thread, generating its own context across messy and conflicting sources and correcting gaps in its own plan.
• Asks clarifying questions when the prompt is ambiguous, asks for help when stuck, and confirms before consequential actions — three behaviours that matter more than a benchmark point when the agent has write access.
• Maps incoming prompts to the right task inside a messy single thread, whether the user is steering a past request or interrupting it.
• Trained to have a better sense of what it can and cannot do, and to signal when it hits a wall "instead of hallucinating outcomes".

Confirming before a consequential action is the least glamorous item on that list and probably the most valuable one.

Source: Meta AI Research, "Introducing Muse Spark 1.3" (September 2, 2026).