Meta’s new Muse Spark model, designed for coding and other agents, introduces a contributor pricing tier that offers roughly a 95% discount in exchange for user data.

Under this model, 1 million input tokens cost $0.10 instead of $1.25, and 1 million output tokens cost $0.20 instead of $4.25.

The move addresses the challenge of obtaining training data for agentic tools—user session data is critical for reinforcement learning, as seen in the capabilities jump of Claude Code between April and October 2025.

However, many large enterprises remain on standard, token-billed plans to avoid sharing proprietary data, a dynamic Meta’s explicit compensation model aims to change.

Princeton professor Arvind Narayanan suggests this could incentivize companies to be more selective about what data is truly proprietary.

The pricing also adds to competitive pressure among frontier labs, following Anthropic’s cached token price cuts and OpenAI’s recent major price reductions.