Key facts
- Meta is offering a pricing model for its Muse Spark AI model that discounts usage for users who share data.
- The contributor pricing model offers discounts averaging 95% on prompts and model outputs.
- Input tokens are priced at 10 cents per million under the contributor model, down from $1.25.
- Output tokens are priced at 20 cents per million under the contributor model, down from $4.25.
- This initiative aims to address Meta's challenges in acquiring sufficient training data for AI development.
Meta is introducing a new pricing strategy for its Muse Spark AI model, which is designed for operating coding and other agents. The company is offering a significant discount, averaging around 95%, to users who agree to share their prompts and model outputs, thereby contributing to the development of future AI models.
Under this 'contributor pricing' model, one million input tokens will cost 10 cents, a steep reduction from the standard $1.25. Similarly, one million output tokens will be priced at 20 cents, down from $4.25. This initiative appears to be Meta's attempt to overcome challenges in acquiring sufficient training data, a crucial element for enhancing AI capabilities. The company previously faced internal criticism for an initiative to track employee computer usage for training purposes.
Experts note that user data is vital for improving agentic tools. Mario Zechner, developer of the open-source harness Pi, highlighted how storing coding agent sessions significantly boosted capabilities. However, acquiring this data is complex, especially for professional workflows outside software engineering. Arvind Narayanan, a Princeton computer science professor, pointed out that large companies often avoid sharing data, preferring more expensive enterprise plans that offer data governance, even if consumer plans are cheaper.
Meta's new pricing framework may incentivize companies to be more discerning about what data is proprietary and what can be shared. This move also comes amid increasing price competition among leading AI labs, with Anthropic and OpenAI recently announcing price reductions for their models.