Meta is offering steep discounts on its Muse Spark agent model to customers who allow their prompts and outputs to contribute to future model development. The pricing makes data-sharing a visible commercial choice rather than placing it only in an opt-out setting.
Under a standard agreement, one million input tokens cost $1.25 and one million output tokens cost $4.25. The contributor tier lowers those prices to $0.10 and $0.20 respectively, an average reduction of roughly 95 percent. Meta describes the option as a way to lower the cost of prototypes, integrations and larger experiments where training on customer data is acceptable.
The offer targets a persistent problem for agent developers: professional workflows generate the traces needed to evaluate and improve agents, but companies often consider those prompts, files and outputs proprietary. Businesses commonly pay more for enterprise services specifically to gain retention controls and IT governance.
The discount therefore needs to be evaluated as payment for data rights, not simply a cheaper model. Teams must decide whether code, documents, tool calls or model responses could contain confidential information before selecting the tier. Clear separation between shareable experiments and sensitive production work will be essential. Meta gains examples of real agent use; customers gain lower prices only when the value and risk of those examples are acceptable to them.