Meta Muse Spark 1.3: Agentic and Coding Model
Meta AI Research released Muse Spark 1.3 on 2 September 2026, saying it improves performance across agentic and coding tasks and is easier to use in real-world settings: it is designed to better sustain longer-horizon work, collaborating with users and juggling multiple workflows in a single long thread; when given an open-ended objective it 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 says the model was trained across a diverse set of harnesses to generalize to various agentic environments, and 1.3 is available in Muse Code and the Meta Model API with max reasoning.
Tool Overview
Features, steps and FAQ below
Features
- ✓ Long-horizon agentic work: Meta says Muse Spark 1.3 is trained to collaborate with users and juggle multiple workflows in a single long thread, and that on open-ended objectives it uses tools to build 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
- ✓ More proactive collaboration: Meta says it asks clarifying questions when prompts are ambiguous, invokes help from the user when stuck and confirms before taking consequential actions, and on long tasks adapts to user preferences by giving frequent updates or working silently in the background
- ✓ Reliable long-form instruction following: Meta says that versus earlier Muse Spark models it follows complex, long-form instructions more reliably across multi-step tasks, preserving detailed requirements without dropping constraints or drifting from the requested workflow
- ✓ Better multitasking and self-awareness: Meta says it more accurately maps incoming prompts to the correct task within messy single-threaded contexts, and is trained to have a better sense of what it can and cannot do and to avoid hallucinating outcomes when it hits hurdles
- ✓ More efficient and safer coding: Meta says that relative to 1.2 it takes fewer unnecessary turns and is less verbose in common engineering workflows, with Meta engineers' comparisons showing ~20% fewer tool calls and ~25% fewer tokens; it also shows stronger adversarial robustness and resistance to prompt injection and better calibration on irreversible actions
How to Use
- Sign up and start building at dev.meta.ai
- Use Muse Spark 1.3 with max reasoning in Muse Code, which can be installed on macOS or Linux
- Call the model programmatically through the Meta Model API and wire it into your agentic and coding workflows
- For evaluation details, see the official Muse Spark 1.3 multimodal evaluation methodology report
FAQ
What is Muse Spark 1.3?
Meta says Muse Spark 1.3 is released by Meta AI Research, improving performance across agentic and coding tasks and becoming easier to use in real-world settings as part of its work toward personal superintelligence.
Where can I use it?
Meta says Muse Spark 1.3 with max reasoning is available in Muse Code and the Meta Model API, and you can start at dev.meta.ai.
How does it collaborate with users?
Meta says it asks clarifying questions when prompts are ambiguous, invokes help when stuck and confirms before consequential actions, and on long tasks gives frequent updates or works silently in the background according to user preferences.
How efficient is its coding?
Meta says that relative to Muse Spark 1.2, 1.3 takes fewer unnecessary turns and is less verbose in common engineering workflows, with Meta engineers' comparisons showing about 20% fewer tool calls and 25% fewer tokens.
How has its safety improved?
Meta says 1.3 improves safety along the axes most relevant to agentic and coding capabilities, with stronger adversarial robustness, better resistance to prompt injection and better calibration on irreversible actions.