Meta Rolls Out Muse Spark 1.3 for Longer Tasks, With Max Reasoning Still Pending
The update reaches Muse Code and Meta Model API with features meant to keep agent-style work on track. The highest reasoning setting remains unavailable until additional safety testing is complete.
Listen to this story
The audio brief
Story brief
3 key pointsMeta’s latest Muse release targets a persistent weakness in agentic software: maintaining goals, context, and constraints when work sprawls across one conversation. Muse Spark 1.3 is now usable through Muse Code and the Meta Model API, with existing reasoning modes enabled, but its highest setting is gated by undisclosed safety testing. The model is designed to ask for clarification, surface blockers, and seek...
- 01
Max reasoning has no announced release date and remains blocked on additional safety testing.
- 02
The model can retain tool-built context, identify planning gaps, and reuse earlier thread requests after interruptions.
- 03
Users can choose frequent progress updates or quieter background execution on longer tasks.
Meta AI Research has started rolling out Muse Spark 1.3, an AI model designed to carry work through longer, messier assignments instead of treating each request as a fresh task. It is available in Muse Code and Meta Model API, while its maximum reasoning mode remains pending further safety testing.
Keeping work together across a long thread
Muse Spark 1.3 is designed to sustain longer-horizon work across several workflows in one thread. For open-ended objectives, Meta says it can use tools to build context from messy or conflicting sources, find gaps in its plan, retain what it learns, and produce a final deliverable.
Meta also says the model more accurately routes incoming prompts to the correct task in cluttered, single-threaded exchanges, including when a user revisits an earlier request or interrupts the current one. The company positions that behavior as part of making the model more useful for agentic and coding tasks.
The model is meant to ask before acting
The update makes user involvement a stated part of its design. Muse Spark 1.3 asks clarifying questions for ambiguous prompts, requests help when stuck, and confirms before consequential actions. For long tasks, it can adapt to a preference for frequent updates or silent background work.
Claims of steadier coding and instruction following
Meta says it trained the model on more long-horizon coding tasks and improved its usability in common engineering workflows. The company also says the model follows complex, long-form instructions more reliably than earlier Muse Spark versions, preserving detailed requirements through multi-step work rather than dropping constraints or drifting from the requested process.
Meta further says Muse Spark 1.3 better recognizes its capabilities, limitations, knowledge gaps, and hurdles. The intended result is a model that flags when it cannot finish work rather than claiming an outcome it did not produce.
Editorial analysis
Our Read
This release puts the model update directly into a coding product that has recently moved beyond beta, making day-to-day workflow behavior more important than a benchmark claim alone. Meta is emphasizing whether the system can keep tasks separate, preserve instructions, ask for help, and pause before consequential actions. That is a practical agent proposition, but one that depends on real use rather than the company’s descriptions. The next meaningful event is the arrival of max reasoning after safety testing, and whether Meta explains how that mode changes the experience for Muse Code and API users.
Sources
- research.meta.aiIntroducing Muse Spark 1.3
