Most AI platforms let users opt out of having their interactions shared with developers for improving future systems. Meta has taken that concept and attached a price tag to it.

With its newly launched Muse Spark model—built for powering coding and other autonomous agents—Meta is offering a steep discount, roughly 95% on average, to users who agree to “contribute” to future model development by sharing their prompts and generated outputs. Under a standard agreement, one million input tokens cost $1.25, but under the contributor pricing tier, that same volume costs just $0.10. Output tokens follow a similar pattern: the standard rate is $4.25 per million, while contributors pay only $0.20 per million.

Meta has struggled to secure training data in the past. An internal initiative launched earlier this year to monitor employees’ computer usage drew widespread criticism and was halted in June. That kind of user data is crucial for making agentic tools function effectively. Yet even as model developers increasingly push these tools beyond software engineering into broader professional settings, their capacity to evaluate and refine them is hampered by the complexity of real-world workflows and the lack of digital footprints left behind.

Arvind Narayanan, a Princeton computer science professor, pointed out that there’s strong evidence large enterprises are reluctant to let their data be used for model training. “They 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. (The main difference between the plans is data retention + enterprise IT governance),” he wrote on social media.

Perhaps recognizing this dynamic, Meta is now offering companies direct financial incentives to share that information. Its pricing guide explains that the contributor tier “lowers the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable.”

Narayanan suggested this approach could push large firms to become more deliberate about distinguishing truly proprietary data from information that could safely be shared with model providers. The framework might also intensify the price competition already heating up among frontier labs. Anthropic’s newest Fable and Mythos models, released yesterday, introduced reduced costs for processing cached tokens, while OpenAI’s latest models saw substantial price cuts at the end of July.