Muse Spark 1.3

Meta's frontier agentic and coding model, and the model consumer Muse runs on, available through the Meta Model API.

Juggles multiple workflows inside a single long thread, proactively corrects its own plan, and asks clarifying questions when a prompt is ambiguous
Uses roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 on coding workflows, by Meta's internal comparison, so the efficiency gain is the headline rather than raw capability
Improved robustness against adversarial input and prompt injection
Benchmarked by Meta against Muse Spark 1.2, GPT 5.6 Sol (max), and Opus 5 (max)
Two API tiers: Standard at $1.25 per million input tokens and $4.25 per million output, the same rate as 1.2; and Contributor at $0.10 per million input and $0.20 per million output
The Contributor discount is paid for with data rather than money: Meta gains rights to use your data for product improvement, and the rate limit drops from 3,000 requests per minute to 100
Available on the Meta Model API, in Muse Code on macOS and Linux, and rolling out to Meta AI, Instagram, and Facebook; it is the model behind consumer Muse
Supersedes Muse Spark 1.2 as the flagship, though Meta announced no formal deprecation of 1.2

Long-running agent and coding work that has to hold constraints across many steps in one thread

Token and tool-call cost is the thing limiting how much agentic work you can run

Teams paying per token for agent workloads, and anyone weighing whether to trade data rights for a cheaper rate

Join the newsletter to get ahead with AI.

I share strategies, new modules, free guides, and tools to get you better at using AI in your work.