Meta priced your training data at a 12x discount.

Meta shipped Muse Spark 1.2 on Aug. 5 with two prices attached to the same model. The standard tier costs $1.25 per million input tokens and $4.25 per million output tokens. A muse-spark-1.2-contributor tier charges $0.10 and $0.20, per Simon Willison's write-up, in exchange for one thing: consenting to let Meta train on your code.

That's roughly 12 times cheaper on input, about 21 times cheaper on output. Other labs bury this kind of trade in terms of service nobody reads twice. Meta put a number on it and let you do the arithmetic yourself.

For a solo builder shipping side projects, the contributor tier is close to free money. No client, no NDA, no reason to say no. For anyone writing code under contract, the standard tier became the only defensible option, and that's worth writing into whatever policy governs which AI tools your team can touch.

Willison's post doesn't say whether the contributor data gets used beyond training Muse Spark itself, or whether Meta will keep offering this split on future models. Until that's answered, the discount is a one-model deal, not a standing policy, and pricing it in as permanent would be a mistake.