AI & Models
Meta discounts its AI model for users who share their data
Meta is offering users of its new Muse Spark model discounts of roughly 95% if they let the company use their prompts and outputs to train future models.
Most AI tools let users opt out of having their usage shared with the model provider for training future versions. Meta has taken the opposite approach with its new Muse Spark model, built for coding and other agents: it is offering an explicit discount that averages out to about 95% for users who “contribute” their prompts and model outputs to future model development.
Under a standard agreement, 1 million input tokens cost $1.25; under the contributor pricing model, the same volume costs just 10 cents. Output tokens are $4.25 per million under the standard price, but 20 cents per million for contributors.
The pricing shift comes after a rockier attempt by Meta to gather training data: an initiative launched earlier this year to track employees’ computer usage drew wide internal criticism and was paused in June. Meta did not respond to a question from TechCrunch about the new pricing model.
User data of this kind is critical to improving agentic tools. Mario Zechner, the developer behind the open source coding harness Pi, told TechCrunch last month that a jump in coding-agent capabilities between April 2025 and October 2025 came in part because a rival coding tool stored users’ coding-agent sessions by default and used them for reinforcement learning. But as model builders increasingly look to deploy agents beyond software engineering, evaluating and improving those tools is limited by the complexity and lack of digital traces in most professional workflows.
Large companies, meanwhile, tend to avoid sharing their data. Arvind Narayanan, a Princeton computer science professor, said on social media: “They stick with token-billed Enterprise plans even though the subscription-based consumer plans like Claude Max and ChatGPT Pro are discounted by 10x-20x or even more! (The main difference between the plans is data retention + enterprise IT governance).” Meta’s pricing guide frames its contributor tier as a way to lower the barrier to entry for prototyping and testing where training on customer data is acceptable — an approach Narayanan suggested could push large companies to be more deliberate about which data is genuinely proprietary and which can be shared.
The pricing move also lands amid broader price competition among frontier labs: Anthropic’s newest Fable and Mythos models, released yesterday, came with lower costs for processing cached tokens, while OpenAI cut prices on its latest models at the end of July.
Why it matters
By directly paying for usage data, Meta is acknowledging that real customer data — not just larger-scale training — is becoming the key bottleneck for improving AI agents.