Meta Offers 95% Muse Spark Discount for AI Training Data

Meta is using a steep pricing incentive to tackle one of the hardest constraints in developing AI agents: a shortage of high-quality data showing how people give instructions, assess outputs and complete multi-step tasks. Under the arrangement for its Muse Spark model, customers can trade access to their prompts and generated responses for lower usage costs. The approach could improve model training, while putting consent, data governance and the commercial value of user interactions under scrutiny.

Meta is offering discounts of as much as 95% on Muse Spark token charges to users who agree that their prompts and model outputs may be used for training. The disclosed details did not specify an exact launch date, the undiscounted token price or how long the offer will run. The move effectively treats lower inference costs as compensation for valuable interaction data, giving Meta a potentially faster route to refine agent behavior and strengthen the next generation of its AI models.