Meta releases Muse Spark 1.3 with about 20% fewer tool calls and 25% fewer tokens than 1.2

Archive entry: this covers an event from September 2, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On September 2, 2026, Meta released Muse Spark 1.3, available with max reasoning on Muse Code and the Meta Model API, according to the Meta AI Research announcement. Meta said it used about 20% fewer tool calls and about 25% fewer tokens than Muse Spark 1.2.

What changed

• It asked clarifying questions and confirmed before consequential actions.
• It improved long-instruction following and multitasking.
• It was more robust against prompt injection and better calibrated on irreversible actions.
• Its context window was 1,048,576 tokens.
• API pricing was $1.25 input, $4.25 output and $0.15 cached input per 1M tokens, and a contributor variant cost $0.10, $0.20 and $0.002 respectively.

What it means for people building with AI

Meta's main agent model became cheaper per task and safer for autonomous use. The reductions in tool calls and tokens meant less work per task, and the clarifying questions, the confirmation before consequential actions and the stronger resistance to prompt injection were the changes behind the safety gain. Because the model was available on both Muse Code and the Meta Model API, builders could use the same model in either place.

Documentation and installation

Meta listed the prices in its pricing and rate limits documentation. Muse Code was installed with curl -fsSL https://dev.meta.ai/install.sh | bash.

Sources

• Introducing Muse Spark 1.3 (Meta AI Research, September 2, 2026, official)
• Pricing and rate limits (Meta (dev.meta.ai docs), official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.