Meta Announces Muse Spark 1.3 for Extended Agent Workflows and Coding

Meta reported the release of Muse Spark 1.3, an AI model designed for long-horizon agentic workflows and code generation that cuts required tokens by 25% compared to its predecessor. Available via API, it costs $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens. The release highlights an industry trend toward enhancing autonomous agent reliability, reducing API token overhead, and optimizing system-level efficiency. Lowering tool call frequency and improving context retention are key factors for deploying cost-effective AI agents in complex software engineering tasks. Muse Spark 1.3 reduces unnecessary code iterations and tool calling requests by approximately 20% while maintaining complex constraint requirements in single long threads. Additionally, it features stronger self-awareness regarding its capabilities, asking clarifying questions when encountering ambiguous user instructions.