Meta Platforms Launches Muse Spark 1.1 AI Coding Model

Meta Platforms released Muse Spark 1.1, an updated artificial intelligence model optimized for advanced coding and agentic workflows. Available to United States developers through a public API preview and integrated into the Meta AI website and application, the model features native multimodal perception for documents, videos, and images. It includes a one-million-token context window and the ability to detect and resolve complex software bugs.

Mark Zuckerberg, the CEO of Meta Platforms, described the pricing for the new model's API as aggressive and attractive, undercutting competitors like OpenAI and Anthropic by offering the model at approximately 25% of their cost. Meta is implementing a pay-as-you-go structure of $1.25 per million input tokens and $4.25 per million output tokens, while providing $20 in free credits for new accounts. This launch marks the first time the company is charging businesses for access to its models via a paid developer tier.

Alexandr Wang, the chief of AI at Meta Platforms, stated that Muse Spark 1.1 is the company's most capable model for coding tasks to date. Developed by Meta Superintelligence Labs, the model is designed to integrate with common developer harnesses to drive adoption. This release follows the rollout of Muse Image and signals a strategic transition toward proprietary models, alongside a new $10 billion data center investment in Canada. While the model shows improvement over its predecessor in vulnerability tasks, evaluation reports indicate it still trails some high-end competitors in offensive security tasks.