Meta has officially entered the paid AI model business with the launch of ‘Muse Spark 1.1’, the first commercial version of its flagship AI reasoning model available to external developers. The model is being offered through the newly launched Meta Model API. Until now, the company mainly focused on releasing open-weight Llama models and providing Meta AI for free across its apps. But with Muse Spark 1.1, the Mark Zuckerberg-led firm is now competing directly with companies like OpenAI, Google and Anthropic in the fast-growing AI API business.
The API is currently available in public preview for developers in the US, and the social media behemoth is offering $20 worth of free API credits before switching to pay-as-you-go pricing. Importantly, pricing is set at $1.25 per million input tokens and $4.25 per million output tokens, positioning the model between lower-cost and premium AI offerings in the market.
Notably, Muse Spark was first introduced earlier this year as the first frontier AI model developed under Meta Superintelligence Labs (MSL), the company’s new AI division formed to speed up the development of advanced AI systems. The new Spark 1.1 version is aimed at developers and businesses, with major improvements in coding, reasoning and long-task execution. According to the firm, the model performs much better at understanding large codebases, finding bugs across multiple files, writing production-ready code and solving complex problems that require multiple reasoning steps. It is also a multimodal model, meaning it can understand text, images, documents and videos, allowing developers to build applications that work with different types of content.
One of the major improvements in Muse Spark 1.1 is its support for agentic AI. The model is designed to plan, reason and carry out longer workflows with less human input. Developers can build AI agents capable of writing and testing code, analyzing documents, carrying out research, using software tools and managing complex workflows. Meta claims that the model also supports multi-agent systems, where multiple AI agents work together on different parts of a problem before combining their results.
Meta has also shared several benchmark improvements for Muse Spark 1.1. The company says the new model achieves state-of-the-art performance on multiple coding and agent benchmarks, outperforming Claude Opus 4.8 and GPT-5.5 on some internal evaluations related to software engineering and AI agents.
The release comes just days after Meta introduced Muse Image, its first proprietary image-generation model, showing that the company is building a complete family of closed AI models covering reasoning, coding, multimodal understanding and image generation.