TLDRocket
Sign in

Meta is paying to peek at how you use their latest AI model

TechCrunch Tim Fernholz Covered by 7 sources

Meta is discounting its new AI model if customers share prompts and outputs. Cheaper access now comes with a trade: let Meta learn from your work.

Based on reporting by TechCrunch, Tim Fernholz — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Meta has started putting a price on something most AI vendors only ask for politely: the right to learn from how people actually use their tools. With Muse Spark, its new model for coding and other agents, companies can choose a much cheaper contributor tier if they agree to share prompts and model outputs for future training.

The difference is not small. Under the standard rate, 1 million input tokens cost $1.25 and 1 million output tokens cost $4.25. Under the contributor setup, those same volumes drop to 10 cents and 20 cents. Meta is effectively offering about a 95% discount in exchange for data that helps it improve the model.

That data matters more for agentic tools than for chatbots that just answer a question and move on. These systems need a trail of real usage to get better, but professional workflows are often messy, private, and hard to measure. Meta has already learned how difficult this can get: an earlier effort to track employee computer usage drew internal backlash and was paused in June.

The company did not answer TechCrunch’s question about the pricing scheme. But the logic is pretty clear. Meta seems to be trying to make data sharing feel less like a hidden tax and more like a line item, while also nudging companies to decide what really has to stay sealed off. That may be enough to loosen some wallets.

It also lands in the middle of a broader price war. Anthropic’s newest Fable and Mythos models were just updated with lower cached-token costs, and OpenAI cut prices at the end of July. The race is no longer just about who has the smartest model. It is about who can make the economics weird enough that companies stop saying no.

My take — AI-written commentary, not fact-checked reporting

This is the part of AI nobody likes to say out loud: the cheapest model is often the one getting fed your work. Meta is being unusually honest about that bargain, which is more respectable than the usual privacy theater, but it also shows how dependent these systems are on other people’s data. The whole market is drifting toward paying users to become unpaid lab assistants, only now with invoices.

Read more about this at: TechCrunch

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.