Meta will pay for sharing how you use Muse Spark
What's the story
Meta has introduced a new pricing model for its latest Muse Spark AI system, which is designed to run coding and other agents. The company is offering an average discount of around 95% to users who are willing to share their prompts and model outputs. This move is aimed at encouraging user participation in the development of future models by sharing their usage data with Meta.
Cost reduction
Contributor pricing model
Under the contributor pricing model, the cost of input tokens has been drastically reduced.
While a million input tokens usually costs $1.25, it now costs just 10 cents under this new model.
Similarly, for output tokens, the standard price is $4.25 per million but it comes down to just 20 cents under the contributor model.
This drastic reduction in costs is aimed at incentivizing users to share their data with Meta.
Data challenges
Challenges in data collection
Despite the new pricing model, Meta has struggled to collect training data.
The company had previously launched an initiative to monitor its employees' computer usage but faced widespread internal criticism and had to pause it in June.
This highlights the challenges tech giants face in collecting user data for improving their AI models, even when they offer incentives like discounted pricing models.
Data significance
Importance of user data
User data is crucial for improving agentic tools, as seen in the significant jump in coding agent capabilities between April and October 2025.
This was largely due to Claude Code storing all coding agent sessions by default and using them for reinforcement learning training.
The case highlights how valuable user data can be in enhancing the performance of AI models.
Data sharing hesitance
Corporate hesitance in data sharing
Princeton computer science professor Arvind Narayanan has observed that large companies are often reluctant to let their data be used for model training.
He noted that these companies prefer token-billed Enterprise plans over discounted subscription-based consumer plans, which offer 10x-20x discounts or more.
The hesitance stems from concerns over data retention and enterprise IT governance, highlighting the complexities of corporate data sharing in AI development.
Data incentive
Explicit compensation for data
In light of these challenges, Meta is now offering companies explicit compensation for their data.
The company's pricing guide states that the contributor tier "lowers the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable."
This could potentially encourage large companies to be more discerning about which data they consider proprietary and which could be shared with model providers.