Meta offers 95% discount on Muse Spark for users who share AI training data

Muse Spark is designed for coding and other software agents. Under the contributor tier, one million input tokens cost $0.10 instead of the standard $1.25, and one million output tokens cost $0.20 instead of $4.25.

TechCrunch said this kind of user data is vital for making agentic tools work better. Mario Zechner, developer of the open-source harness Pi, told the publication that coding-agent capabilities jumped sharply between April 2025 and October 2025 in part because Claude Code stored sessions by default and used them for reinforcement learning training.

Meta has had difficulty obtaining training data. A company initiative to track employees' computer use, launched earlier this year, drew broad internal criticism and was paused in June, the report said. Meta did not respond to TechCrunch's question about the contributor pricing model.

Princeton computer science professor Arvind Narayanan said there is good evidence that large companies do not want their data used for model training. He noted that such companies stick with token-billed enterprise plans even though consumer plans such as Claude Max and ChatGPT Pro are discounted by 10 to 20 times, with the main differences being data retention and enterprise IT governance.

Meta's pricing guide says the contributor tier "lowers the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable." Narayanan suggested the plan could push companies to define more carefully which data is proprietary and which can be shared with model providers.

The model pricing comes as competition among AI labs intensifies. Anthropic released its newest Fable and Mythos models a day earlier with lower costs for processing cached tokens, while OpenAI cut prices on its latest models in late July, TechCrunch noted.