Meta cuts Muse Spark prices 95% for users who share their coding data

Meta is offering steep discounts on its new Muse Spark coding model to users willing to let the company train future AI systems on their prompts and outputs, according to TechCrunch. The move breaks from the usual opt-out approach, where AI companies let users decline data sharing at no extra cost, and instead puts an explicit price on that choice.

Under Meta’s standard pricing, Muse Spark costs $1.25 per million input tokens and $4.25 per million output tokens. A token is a small chunk of text an AI model reads or writes. Users who agree to contribute their prompts and outputs for training pay just 10 cents per million input tokens and 20 cents per million output tokens, a discount of roughly 95%. Meta’s pricing guide describes the contributor tier as a way to lower “the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable.” Meta did not respond to a question from TechCrunch about the new pricing model.

The offer follows a rockier attempt by Meta to gather training data. Earlier this year the company launched a program tracking employees’ computer usage, then paused it in June after internal criticism, TechCrunch reported.

Real-world usage data has become central to improving coding agents, AI systems that carry out multistep programming tasks on their own. Mario Zechner, developer of the open-source coding tool Pi, told TechCrunch last month that Anthropic’s Claude Code stored users’ coding sessions by default and used them for reinforcement learning, a training method that improves a model through trial and feedback. Zechner said that data explains “the big jump in [coding agent] capabilities between April 2025 and October 2025.”

Big companies already avoid data-sharing plans

Princeton computer science professor Arvind Narayanan pointed out that large companies already show a preference for keeping their data private, even at a steep cost. He noted on social media that big customers “stick with token-billed Enterprise plans even though the subscription-based consumer plans like Claude Max and ChatGPT Pro are discounted by 10x-20x or even more,” largely because enterprise plans include stronger data retention controls and IT governance. Narayanan suggested Meta’s explicit pricing split could push large companies to sort more carefully which data is genuinely proprietary and which they are comfortable sharing.

The discount also lands amid broader price competition among AI labs. Anthropic’s newest Fable and Mythos models, released the day before Meta’s announcement, came with lower costs for processing cached tokens, while OpenAI cut prices on its latest models at the end of July.

What happens next depends on whether other model makers follow Meta in pricing data access explicitly, or keep opt-out sharing bundled into standard plans without a separate price attached.