Updated
Updated · TechCrunch · Sep 3
Meta Slashes Muse Spark Prices 95% for Users Who Share AI Data
Updated
Updated · TechCrunch · Sep 3

Meta Slashes Muse Spark Prices 95% for Users Who Share AI Data

3 articles · Updated · TechCrunch · Sep 3

Summary

  • $0.10 per million input tokens and $0.20 per million output tokens are available under Meta's new contributor tier, versus standard rates of $1.25 and $4.25, if customers let Muse Spark use prompts and outputs for future training.
  • Meta is effectively paying for data because agentic AI systems need real usage traces to improve, and the company has already struggled to gather them after an employee computer-tracking initiative drew backlash and was paused in June.
  • Large companies have often paid more for enterprise plans to keep data out of model training, Princeton professor Arvind Narayanan noted, suggesting Meta's discount could push customers to separate truly proprietary data from information they can share.
  • The move also sharpens AI price competition a day after Anthropic cut some token-processing costs and weeks after OpenAI rolled out major model price reductions, even as Meta earlier upgraded Muse Spark 1.3 without changing API prices.

Insights

Could Meta's highly efficient coding AI actually cause enterprise software development budgets to skyrocket due to the Jevons Paradox?
Are raw benchmark scores blinding developers to the massive hidden orchestration costs required to make these coding agents truly autonomous?
With autonomous agents editing multiple files, how can enterprises prevent malicious prompt injections from hijacking their software supply chain?