Muse Spark 1.1

A multimodal reasoning model from Meta built for agent work, now in the Meta AI app and open to developers through a new API in public preview.

Handles images, video, and PDFs, with a 1 million token context window
Built for tool use and computer automation across multiple applications, and for delegating to and orchestrating other agents
The gains over the prior version are in tool and computer use, coding, and multimodal understanding
Judges when to automate a step versus interact with an interface directly
Reachable in Thinking mode in the Meta AI app and at meta.ai, and through the new Meta Model API in public preview, which is OpenAI-compatible