Meta is offering substantially lower prices for its Muse Spark AI model to customers who allow their prompts and model outputs to be used to develop future models. Under its contributor pricing, input tokens cost $0.10 per million and output tokens cost $0.20 per million, compared with standard prices of $1.25 and $4.25 respectively.
The difference works out to an average discount of roughly 95%, depending on the mix of input and output tokens. Meta describes the lower-cost option as a way to reduce the cost of prototyping, integration testing, and larger experiments when customers are willing to contribute their data.
Meta Links Lower Prices to Model Training Data
Muse Spark is designed for coding and agentic workloads where AI systems can use tools and complete longer sequences of tasks. Meta recently detailed the latest model in its Muse Spark announcement, describing improvements to coding, long-running tasks, tool use, and multi-step workflows.
The contributor plan gives Meta access to prompts and model outputs that can help improve future systems. Such interaction data can be useful when training agents because it shows how models perform during real-world tasks rather than only on prepared datasets.
The pricing difference is substantial. One million input tokens cost 92% less under the contributor plan, while one million output tokens cost about 95% less.
Meta’s approach follows challenges in obtaining data that reflects real computer use. Earlier this year, the company paused an internal program that monitored employee computer activity after workers raised concerns about data security and privacy.
Agent Training Creates Demand for Real-World Usage Data
Developers have argued that records of real agent interactions can significantly improve AI systems. Mario Zechner, developer of the open-source Pi agent harness, said access to coding sessions helped drive rapid improvements in coding agents by providing material for reinforcement learning.
Companies, however, have often paid more for enterprise AI products that provide stronger data-retention controls. Princeton computer science professor Arvind Narayanan noted that businesses continue using token-billed enterprise plans despite cheaper consumer subscriptions, with privacy and governance among the important differences.
Meta’s pricing model offers a direct financial incentive for organizations willing to share some of their AI usage. It could also encourage companies to distinguish between sensitive proprietary information and workflows they are comfortable contributing to model development.
The pricing comes as AI providers compete more aggressively on model costs. Anthropic recently reduced some token-processing prices for its Fable and Mythos models, while OpenAI introduced substantial price reductions for its models in July.
Meta has not publicly provided additional details about how widely customers are choosing the contributor pricing option.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.