Meta introduced Muse Spark 1.1, a multimodal reasoning model for agentic tasks, and opened a public preview of the new Meta Model API. The muse release gives developers a new route to code, tool use, and multimodal work through API access.
Muse Spark 1.1 and Meta Model API
Muse Spark 1.1 is the latest model from Meta Superintelligence Labs. Meta says the model has major gains in tool and computer use, coding, and multimodal understanding, which matters for teams trying to automate work that moves between text, tools, and code instead of staying inside one prompt.
The model is available in Thinking mode in the Meta AI app and on meta.ai. That gives individual users a way to try the model directly, while developers can reach it through the new Meta Model API for application work.
Thinking mode on meta.ai
Muse Spark 1.1 zero-shot generalizes to new native tools, MCP servers, and custom skills. In practice, zero-shot use means a developer can ask the model to work with a capability it has not been separately trained on for that exact setup, then see whether it can still follow the tool shape well enough to act on it.
The model also actively manages a context window of 1 million tokens. That is the memory span it can keep in view while it works, which gives it more room to track long codebases, longer task chains, or multiple references without dropping earlier details.
OpenCode and agentic tasks
Meta says Muse Spark 1.1 was trained to orchestrate multi-agent systems to optimize end-to-end latency. It can gather context, make a plan, and delegate execution across parallel subagents, which is the kind of workflow developers use when one model has to split a job instead of handling every step in a single pass.
In OpenCode, the model builds a chat web app, takes automated screenshots to identify user-visible failures, traces issues back to relevant code, implements fixes, and validates the changes. That workflow points to a concrete use case for developers who want an assistant that can move from diagnosis to patching without switching tools after every step.
The catch is that the release text treats Muse Spark 1.1 as a major upgrade, but it does not publish the underlying evaluation report details behind that performance claim. Developers get the preview now, but the practical question is whether the public API behaves as well outside Meta’s own demonstrations as it does inside them.
