Meta Platforms has announced the release of Muse Spark 1.3, the latest version of its advanced AI model.
Muse Spark 1.3 is designed to improve performance in both agentic and coding tasks.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Meta has highlighted enhancements based on feedback from earlier use of Muse Code and the Meta Model API, aiming to make the system easier to use for practical applications.
According to Meta, the new model has been trained to better handle longer-term tasks that require collaboration with users and management of multiple workflows.
It can generate its own context from a variety of sources when pursuing open-ended objectives and has been equipped to ask clarifying questions if prompts are unclear.
The company also stated that Muse Spark 1.3 seeks user help if it encounters problems and requests confirmation before making significant changes.
Meta said that the updated system follows complex, long-form instructions more accurately than previous versions.
The model’s ability to manage multi-step tasks and maintain adherence to specific requirements has been strengthened.
Meta, in its announcement, said: “We’re excited to release Muse Spark 1.3, which delivers improved performance across agentic and coding tasks.”
On the technical side, the coding performance has been upgraded, with the company reporting that the new version is faster and uses fewer resources than Muse Spark 1.2.
Meta engineers found that it reduced tool usage by around 20% and token usage by around 25%, while also offering a less verbose and cleaner coding style.
The company has also introduced safety improvements, citing increased resistance to adversarial inputs and a stronger ability to recognise potentially irreversible actions.
Muse Spark 1.3 is now available through Muse Code and the Meta Model API. It will maintain the same pricing as the previous Muse Spark 1.2 release.
Last month, Meta launched Muse Code, a coding agent accessible via the terminal and powered by the company’s AI model, Muse Spark 1.2.
The agent is intended to support software developers in writing and debugging code, with a particular focus on large and complex projects.
Muse Code enables code generation and accuracy verification, and it manages multiple persistent background sub-agents operating throughout each developer session.
