Muse Spark 1.3 is the latest AI model from Meta’s Muse family, released September 2, 2026. It is built for professional use cases involving extended workflows: coding, agentic tasks (where the model manages multiple sub-tasks over time), long documents, multimodal inputs (text, images, video, PDF), and more. It is available through Muse Code (a coding agent) and the Meta Model API. The model introduces more reliable instruction following, improved safety, greater efficiency, and better handling of complex, long-horizon work.
Key Features
Long-horizon agentic workflows: Muse Spark 1.3 holds together longer threads of work. It maintains coherence amid messy or conflicting inputs, fills in gaps proactively, asks clarifying questions when prompts are ambiguous, and confirms before taking consequential actions.
Improved coding efficiency: Compared to its predecessor (Muse Spark 1.2), 1.3 delivers roughly 20% fewer tool calls and about 25% fewer tokens in engineering benchmarks. It’s less verbose and more cleanly styled in code output.
Massive context window & multimodal input: Supports up to ~1,048,576 tokens (1 million tokens), and accepts text, image, video, and document (PDF) inputs. It also supports output in structured formats, tool invocation, and search capabilities.
Safety and reliability enhancements: Better resistance to adversarial prompts and “prompt injection,” improved calibration for recognizing irreversible actions, ability to admit what it doesn’t know rather than hallucinate outcomes.
Pricing
Public pricing for Muse Spark 1.3 is well defined through the Meta Model API and Muse Code. There are two main price tiers:
Standard tier: $1.25 per 1 million input tokens; $4.25 per 1 million output tokens; cached inputs cost $0.15 per million tokens. Requires no consent for Meta to use prompts and outputs for training.
Contributor tier: Much lower rates — $0.10 per 1 million input tokens; $0.20 per 1 million output tokens; cached input at $0.002 per million tokens — in exchange for permission for Meta to use your prompts and generated outputs in its training pipeline. Contributor also has stricter rate limits.
Pricing for “max reasoning” mode (a more intensive reasoning effort) is part of the same model; no separate SKU is listed. Standard pricing applies.
Who is it for?
Muse Spark 1.3 is aimed at organizations or roles who need AI to manage complex, ongoing tasks rather than simple Q&A. It is well suited for:
Software engineering teams working on large codebases or needing AI assistance with programming, debugging, long-term project management.
Data analysts, researchers or policy professionals handling large documents, reports, technical specifications, or analysis workflows requiring consistency and preserving constraints over many steps.
Businesses building AI agents that interact with multiple inputs and tools over long time frames — such as automation platforms, customer support flows, or workflow orchestration systems.
Enterprises or startups mindful of budget who can consider using the Contributor tier and accept that data shared may be used for model training—this trade-off can lead to significantly lower cost for heavy usage.
It is less suitable for organizations needing fully local deployment with open-source weights because Muse Spark remains proprietary and hosted by Meta. Also, if privacy of prompts and outputs is critical, the Standard tier offers more control but at higher cost.
Final thoughts
Muse Spark 1.3 represents a forward step for Meta in enterprise-grade AI. It combines a large context window, efficiency gains, stronger safety, and multimodal reasoning in a single hosted model. For businesses looking to deploy AI in agentic workflows or demanding coding tasks, it offers a competitive option. Cost trade-offs via Contributor vs Standard tiers are meaningful, and getting actual performance in your workload will be essential. For decision-makers, the key will be matching usage demands (volume, privacy, rate limits) against the pricing model, and weighing whether the advantages in Muse Spark 1.3 over earlier versions or competing models justify adoption.
Visit the official website for more.
Keep up to date with our stories on LinkedIn, Twitter, Facebook and Instagram.