Meta on September 2, 2026 released Muse Spark 1.3, the latest model from Meta Superintelligence Labs, making it available the same day in Muse Code and in Meta Model API. Meta said the update delivers improved performance across agentic and coding tasks and is easier to use in real-world settings, drawing on what the company described as months of broad adoption of Muse Code and Meta Model API.
According to the Meta AI Research announcement, the previously available reasoning modes are available at launch, while a “max reasoning” mode will come shortly after Meta finishes additional safety testing. Meta pointed readers to an accompanying evaluation report for benchmark details.
Agentic Workflows
Meta said Muse Spark 1.3 is designed to better sustain longer-horizon work by collaborating with users and juggling multiple workflows in a single, long thread. When given an open-ended objective, the company said, the model uses tools to generate its own context across messy and conflicting sources, proactively corrects gaps in its plan, and keeps track of what it has learned to produce a final deliverable. Meta said it trained the model across a diverse set of harnesses to generalize to various agentic environments.
The company said it trained Muse Spark 1.3 to more actively collaborate with the user: the model asks clarifying questions when prompts are ambiguous, invokes help from the user when stuck, and confirms before taking consequential actions. On long tasks, Meta said, it adapts to user preferences, either providing frequent updates or working silently in the background.
Meta said Muse Spark 1.3 follows complex, long-form instructions more reliably than earlier Muse Spark models, and that across multi-step tasks it is better at preserving detailed requirements without dropping constraints or drifting from the requested workflow.
The company also said it improved the model’s multitasking capabilities. As an example, Meta said Muse Spark 1.3 now more accurately maps incoming prompts to the correct task within messy, single-threaded contexts, regardless of whether the user is steering past requests or interrupting them. Meta added that the model has better awareness of its own capabilities and limitations, saying it was trained to have a better sense of what it can and cannot do, what it knows and does not know, and when it hits hurdles instead of hallucinating outcomes.
Coding
Meta said Muse Spark 1.3 was trained on more long-horizon coding tasks and shows improved usability in common engineering workflows. Relative to Muse Spark 1.2, the company said, the new model takes fewer turns where not needed and is less verbose, while having a cleaner overall coding style. In comparisons by Meta engineers, the company said, the model proved to be significantly faster and more efficient, using roughly 20% fewer tool calls and roughly 25% fewer tokens.
Safety and Availability
Meta said it improved safety along several axes it described as most relevant to agentic and coding capabilities. The company said Muse Spark 1.3 shows stronger adversarial robustness, with improved resistance to adversarial inputs and prompt injections. On complex agentic tasks, Meta said, the model has better calibration on what constitutes irreversible actions and proceeds accordingly. The company said these changes reflect better discretion and judgment in long-horizon agentic tasks.
Muse Spark 1.3 is available today in Muse Code and in Meta Model API, Meta said.
Prior State and Roadmap
Muse Spark 1.3 succeeds Muse Spark 1.2, which Meta released on August 5, 2026 alongside Muse Code, a terminal coding agent. In that earlier Muse Code and Muse Spark 1.2 announcement, Meta described Muse Spark 1.2 as a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. Meta said it significantly scaled up training compute on coding tasks for that release while expanding training environment diversity, and that the model was co-trained with Muse Code.
Looking forward, Meta said its roadmap includes bigger models, the Muse Spark open weights release, and more, without providing dates.
