Meta has shifted its strategy from leading open-source AI development to competing primarily on price. The company released Muse Spark 1.2, a new model paired with Muse Code, a coding agent designed to resume work after system failures without losing context.
The pricing reflects Meta's new market position. The cheapest tier costs just 20 cents per million output tokens, but comes with a significant tradeoff: users must consent to have their data used for training. This represents a fundamental change for Meta, which previously earned credibility by releasing open-weight models like Llama that shaped the broader AI industry. Now the company positions itself as a budget alternative rather than a performance leader.
Meta faces real benchmarking gaps that the pricing strategy attempts to sidestep. Rather than compete directly with frontier models from OpenAI or Anthropic on capability metrics, Meta targets price-sensitive customers willing to trade performance for cost. Developers and companies operating at scale find this approach appealing, especially when working on less demanding tasks where state-of-the-art reasoning matters less than affordability.
The move reveals how the AI market is consolidating around different tiers. Frontier labs push capabilities forward at premium prices. Open-source projects prioritize accessibility and community control. Meta now occupies a middle ground: semi-open models sold at discount rates with data collection attached. This generates revenue while maintaining the brand association with openness that made Llama successful.
The data collection clause changes the calculus for enterprise customers who handle sensitive information. A 20-cent price point becomes expensive if it means training data exposure. Meta essentially offers a choice: pay more for privacy or accept data sharing for deep discounts.
The release of Muse Code suggests Meta also recognizes coding as a competitive domain where agents matter as much as base models. By bundling task-specific tools with cheaper models, Meta attempts to deliver