Most AI tools allow users to choose whether to share their data with the developers, which is used to improve future versions. Meta, the company behind platforms like Facebook and Instagram, has introduced a new pricing model for its Muse Spark AI model, which is designed for tasks like coding and other agent-based applications. This model offers significant discounts to users who agree to share their prompts and the model's outputs with Meta. The discount is substantial, averaging around 95%. For example, 1 million input tokens normally cost $1.25, but under the contributor pricing model, they cost only 10 cents. Similarly, output tokens, which typically cost $4.25 per million, now cost just 20 cents under the new model. Meta has had difficulty gathering the data needed to train its AI models. Earlier this year, the company launched an initiative to track the computer usage of its employees, but the project faced internal criticism and was paused in June. When asked about the new pricing model, Meta did not provide a response. User data is crucial for improving AI tools, especially those designed to perform complex tasks like coding. Mario Zechner, a developer of an open-source AI tool called Pi, mentioned that the significant improvements in coding agent capabilities between April 2025 and October 2025 were partly due to the default data collection practices of another AI model, Claude Code. As AI tools are increasingly used in areas beyond software engineering, evaluating and improving them becomes more challenging. Professional workflows often involve complex processes that leave few digital traces, making it difficult for developers to understand how these tools are used. Arvind Narayanan, a computer science professor at Princeton, pointed out that many large companies are reluctant to share their data for model training. He noted that despite the availability of cheaper consumer plans, such as Claude Max and ChatGPT Pro, which can be up to 20 times less expensive, large companies often stick with more expensive, subscription-based enterprise plans. The key difference, he said, is data retention and IT governance. In response to these dynamics, Meta is offering companies financial incentives to share their data. According to Meta’s pricing guide, the contributor tier lowers the cost barrier for experimenting and testing AI tools when using data that is acceptable to share. Narayanan suggested this could encourage companies to be more discerning about what data is truly proprietary and what can be shared with AI model providers. This approach might also increase competition among leading AI research labs. For example, Anthropic recently released new models with reduced costs for cached tokens, and OpenAI also introduced significant price cuts for its latest models in July.