Meta is offering users of its new Muse Spark model an average discount of about 95% if they share prompts and model outputs to help develop future models. The model is designed to operate coding and other agents. One million input tokens, units of model text, cost $1.25 under the standard agreement versus $0.10 under the contributor tier, while output tokens cost $4.25 versus $0.20 per million. The pricing comes as Meta faces challenges obtaining training data, including criticism that paused an employee computer-usage tracking initiative launched earlier this year. Developers say retained coding-agent sessions can improve reinforcement learning, while companies have generally resisted allowing their data to be used for model training. Meta's pricing guide says the contributor tier is intended to reduce costs for prototyping, integration testing and experiments where customers accept training on their data. The approach could encourage companies to distinguish proprietary information from data they can share, while adding to price competition with Anthropic and OpenAI.