Meta opens Muse Spark 1.1 through its new Model API
The story in five sentences
- Meta has introduced the multimodal agent model Muse Spark 1.1.
- Developers receive access for the first time through a public preview of the Meta Model API.
- The model is also available in Meta AI’s Thinking mode and on meta.ai.
- Meta reports a one-million-token context window and capabilities for tools, computer control, and coding.
- Most performance and safety claims rely on Meta’s own evaluations.
Full explanation
Meta introduced Muse Spark 1.1 on July 9 and identifies MSL for the model. Developers can access it through the public preview of the Meta Model API, while users can reach it through Thinking mode in Meta AI and on meta.ai. The announcement does not provide pricing, usage quotas, or a date for general availability, so conditions beyond the preview remain unspecified.
The described architecture uses a main agent to gather context, create a plan, and delegate work. Subagents receive a defined scope and can escalate when necessary. Meta reports a context window of one million tokens. The model is presented as capable of using tools, controlling computers, running scripts, processing visual and audio material, and supporting coding tasks.
Meta says it evaluated Muse Spark 1.1 under its Advanced AI Scaling Framework. The company states that the model remains within its safety boundaries for chemical, biological, and cyber risks as well as loss-of-control scenarios. It also claims greater robustness and fewer hallucinations. Those conclusions and a cited internal coding comparison are predominantly based on Meta’s own evaluations rather than independent confirmation.