Muse Spark 1.3

// Overview

Muse Spark 1.3 is Meta's update to Muse Spark 1.2, released on September 2, 2026 and rolled out the same day in Muse Code and the Meta Model API. Meta describes it as an improvement across agentic and coding tasks, shaped by what the team learned from months of Muse Code and Meta Model API usage.

The headline change is efficiency rather than raw capability. In side-by-side comparisons by Meta engineers, Muse Spark 1.3 proved significantly faster and more efficient than Muse Spark 1.2, using about 20% fewer tool calls and about 25% fewer tokens. Meta says the model is less verbose, takes fewer turns where they are not needed, and has a cleaner overall coding style.

For agentic work, Meta trained the model across a diverse set of harnesses so behaviour generalizes between agent environments. Given an open-ended objective it uses tools to build its own context from messy and conflicting sources, corrects gaps in its plan, and tracks what it has learned. It is also trained to ask clarifying questions on ambiguous prompts, ask the user for help when stuck, and confirm before taking consequential actions.

Muse Spark 1.3 keeps the 1M-token context window and the standard Meta Model API price of $1.25 per million input tokens and $4.25 per million output tokens. Reasoning modes that were already available shipped on day one; Meta says max reasoning follows after additional safety testing. On safety, Meta reports stronger resistance to adversarial inputs and prompt injections, plus better calibration on which actions are irreversible.

// Pricing

Standard Meta Model API tier. A muse-spark-1.3-contributor tier costs $0.10 input / $0.002 cached / $0.20 output per 1M tokens in exchange for Meta using the data to improve its products.

Pricing source ↗

// Strengths

• Efficiency: about 20% fewer tool calls and about 25% fewer tokens than Muse Spark 1.2 in Meta's own engineer comparisons
• Long-horizon agentic work: trained across many agent harnesses, builds its own context from conflicting sources, and self-corrects gaps in its plan
• Collaborative behaviour: asks clarifying questions on ambiguous prompts and confirms before consequential actions
• 1M-token context window with native multimodal input (video, images and documents)
• Unchanged pricing at $1.25/$4.25 per million tokens, plus a $0.10/$0.20 contributor tier

// Best for

• Reach for it in Muse Code for long-horizon software work where token spend on a single long thread is the real cost driver.
• Reach for it for agent runs that must survive messy, conflicting inputs and multiple interleaved tasks in one thread.
• Reach for it over Muse Spark 1.2 when the job is unchanged but the bill matters — same price per token, fewer tokens used.

// How to access

// Muse Spark — every version

The full lineage of the Muse Spark line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.