Meta Platforms on Thursday made developer access available for Muse Spark and introduced an upgraded iteration, Muse Spark 1.1, moving the company's superintelligence team further into the commercial AI arena. The company described Muse Spark 1.1 as its most capable model to date for tasks involving real-world coding and agentic behavior, framing the work as part of its broader goal to deliver "personal superintelligence."
According to Meta, the updated model can perform a range of complex activities: writing and debugging code, operating software and external tools, interpreting text, images and video, and completing multi-step tasks with reduced human oversight. Those capabilities are central to the company’s pitch about the model’s utility for developers and end users alike.
Muse Spark was first introduced in April as the initial text and reasoning model produced by the superintelligence team that Meta assembled last year. At that time, Meta had been testing an Application Programming Interface in a private preview with partners. The API functions as a connector enabling developers to integrate the model’s capabilities into their own applications and workflows.
Starting now, developers located in the United States can access Muse Spark through a public preview on the Meta Model API. The preview environment allows developers to test prompt designs, compare model outputs, and build prototype integrations. Those who register for the API are provided with $20 in free credits to experiment before transitioning to pay-as-you-go billing.
Meta set the API pricing at $1.25 per million input tokens and $4.25 per million output tokens. The company positioned these rates above OpenAI’s entry-level GPT-5 mini and Anthropic’s Claude Haiku 4.5, while remaining below Anthropic’s higher-end Claude Sonnet 4.6.
The new Muse Spark 1.1 model is available in Thinking mode within the Meta AI app and on Meta’s website. Meta also indicated that Muse Spark 1.1 is expected to replace existing Llama models that currently power chatbots across WhatsApp, Instagram, Facebook and the company’s smart glasses.
This release follows a separate company announcement earlier in the week that expanded generative AI tools across Meta’s apps through the rollout of Muse Image, described as the first image-generation model from Meta Superintelligence Labs.
Context and implications
The public preview and pricing choices position Meta to compete directly with established AI providers by offering developer access to a more capable model while deploying the technology across its consumer-facing products. The availability of Muse Spark 1.1 in app and web environments signals an intention to integrate the model broadly within Meta’s existing services.