Meta releases Muse Spark 1.3

Meta released Muse Spark 1.3 on September 2, saying the update improves agentic and coding performance while making the model easier to use in real-world settings. The company said the release draws on “months of broad adoption” of Muse Code and the Meta Model API, and is aimed at long-horizon workflows, collaboration with users and lower hallucination rates. Meta described the model as part of its push toward “personal superintelligence.” The company said Muse Spark 1.3 is available in Muse Code and in the Meta Model API. In a post from Meta’s AI account on X, the company highlighted better support for multi-step work, more clarifying questions, stronger self-calibration and efficiency gains in workflows. ### What did Meta say changed in Muse Spark 1.3? (research.meta.ai) Meta said the new model is designed to behave more usefully in practical assistant settings, not just score better on headline benchmarks. In its September 2 research post, the company said Muse Spark 1.3 improves long-horizon workflow support and is better at working with users by asking clarifying questions and flagging issues during tasks. (research.meta.ai) The X post from AI at Meta said Muse Spark 1.3 also improves self-calibration to reduce hallucinations. Meta framed the release around day-to-day assistant behavior and smoother execution across multi-step tasks. (research.meta.ai) ### What numbers is Meta using to make its case? Meta said Muse Spark 1.3 reduces tool calls by about 20% and token use by roughly 25% in workflows. Those figures were cited in the company’s social post announcing the release and were presented as efficiency gains tied to practical usage. (siliconangle.com) Meta’s research post said the model also delivers improved performance across agentic and coding tasks. The company did not present the release primarily as a leaderboard push; instead, it said the changes reflect lessons from usage in Muse Code and the Meta Model API over recent months. ### Where does Muse Spark 1.3 sit in Meta’s broader product rollout? (siliconangle.com) Meta introduced the Muse Spark family in April, calling it the first model from Meta Superintelligence Labs built for Meta’s products. At the time, Meta said Muse Spark would power a smarter Meta AI and eventually support features tied to recommendations and shared content across Instagram, Facebook and Threads. (research.meta.ai) In July, Meta said Muse Spark 1.1 was powering new Meta AI features that could plan tasks, connect to email and calendar apps, create slides and handle actions on a user’s behalf. That product framing helps explain why Meta is now emphasizing workflow reliability, user collaboration and reduced hallucinations in version 1.3. (about.fb.com) ### Is Meta positioning this as a benchmark race or a product update? Meta’s September 2 post said Muse Spark 1.3 is “smarter and more practically useful,” language that put product behavior at the center of the release. The company said the model was shaped by real adoption data from Muse Code and the Meta Model API rather than by benchmark performance alone. (about.fb.com) Artificial Analysis said on September 2 that Muse Spark 1.3 was Meta’s fourth Muse Spark release in five months and described the model as reaching the frontier on its intelligence index. That outside assessment focused on competitive standing, while Meta’s own release emphasized usability in deployed assistant workflows. (research.meta.ai) ### What comes next for users and developers? Muse Spark 1.3 is already rolling out in Muse Code and the Meta Model API, according to Meta’s release on September 2. Meta Connect 2026 is scheduled for September 23-24 in Menlo Park, California, where the company has said it will preview what is coming next for its computing platforms. (research.meta.ai) (artificialanalysis.ai)