Meta releases Muse Spark 1.3 with a cheaper tier that trains on your prompts

Meta released Muse Spark 1.3 on September 2, 2026, announced by Mark Zuckerberg on X rather than on a Meta page, and available through Muse Code and the Meta Model API. Axios reports a standard tier at $1.25 per million input tokens and $4.25 per million output, alongside a cheaper contributor tier priced in exchange for Meta training on submitted prompts and completions. It scores 48 on the Artificial Analysis Intelligence Index v4.3, the highest of any Meta model.

Why it matters

The contributor tier makes the trade explicit, which is unusual: pay less, and your prompts and outputs become training data. That is a procurement question and a client-confidentiality question long before it is a price question.

Who should care

Agencies and consultancies handling client data, and anyone comparing per-token prices

What you can do

If the contributor tier looks tempting, check your client contracts and DPAs before enabling it. The standard tier carries no such term.