Meta announced in September 2026 that it is offering steep discounts on its new Muse Spark AI model — up to 95% off standard pricing — for users who agree to share their prompts and model outputs to help train future versions of the model.
Under the standard pricing structure, one million input tokens costs $1.25 and one million output tokens costs $4.25. Users who opt into the “contributor” pricing tier pay just $0.10 per million input tokens and $0.20 per million output tokens. Meta’s pricing guide describes the contributor tier as lowering “the barrier to entry for prototyping, testing integrations, and scaling experiments where training on your data is acceptable.”
Muse Spark is designed for coding and other agentic applications. The move comes after Meta faced internal pushback over a separate initiative to track employee computer usage for training data purposes, which was paused in June 2026 following criticism from staff.
The demand for real-world usage data is tied directly to improving agentic AI tools. Mario Zechner, developer of the open source harness Pi, told TechCrunch that a significant jump in coding agent capabilities between April and October 2025 was driven by Claude Code storing user sessions by default for reinforcement learning training.
Arvind Narayanan, a Princeton computer science professor, noted that large companies tend to avoid sharing data with model providers, often choosing token-billed enterprise plans over cheaper subscription plans specifically because enterprise plans offer data retention controls and IT governance. He suggested Meta’s explicit compensation model could encourage companies to more carefully distinguish between proprietary data and data that could be shared.
The pricing shift also arrives amid broader cost competition among AI labs. Anthropic released its Fable and Mythos models the day prior with reduced cached token pricing, and OpenAI cut prices on its latest models in late July 2026.
Source: TechCrunch