Muse Spark 1.1: Meta's Multimodal Reasoning Model for Agentic Tasks

Introduction to Muse Spark 1.1

Meta Superintelligence Labs has released Muse Spark 1.1, a multimodal reasoning model built for agentic tasks, alongside a public preview of the Meta Model API. This model is designed to think before answering and has an adjustable reasoning effort per request. The context window is 1,000,000 tokens, as listed in the Meta Model API docs.

Key Features of Muse Spark 1.1

Muse Spark 1.1 leads the tool-use and tool-augmented reasoning evaluations, according to Meta's launch table. It has a million-token context window that is actively compacted by the model itself, remembers actions, retrieves information from earlier work, and compacts what it keeps. The model also delegates execution across parallel subagents and has zero-shot generalization to new native tools, MCP servers, and custom skills.

Orchestration Behavior

The model's orchestration behavior explains its tool-use results. As a main agent, it gathers context, plans, and delegates execution across parallel subagents. As a subagent, it adheres to its job, understands available tools, and escalates back when needed.

Access and Pricing

Consumers can access Muse Spark 1.1 for free in 'Thinking' mode in the Meta AI app and on meta.ai. Developers pay $1.25 per million input tokens and $4.25 per million output tokens, with new accounts receiving $20 in free credits. The public preview is currently US-only, with no EU access yet.

Use Cases and Demos

Muse Spark 1.1 has been demonstrated in various use cases, including a Facebook Marketplace demo where the model takes smartphone video, extracts useful photos, reasons about the product, and operates a browser to publish the listing. Another demo shows the model building a chat web app, taking automated screenshots, tracing failures back to code, and validating fixes.

Strengths and Weaknesses

Muse Spark 1.1 leads Meta's reported tool-use and tool-augmented reasoning evaluations but trails Opus 4.8 and GPT-5.5 on coding. The model has a closed weights architecture, which means no local deployment and no fine-tuning. The public preview is US-only, and pricing can change.

Conclusion and Next Steps

The release of Muse Spark 1.1 marks a significant development in multimodal reasoning models for agentic tasks. With its adjustable reasoning effort, million-token context window, and zero-shot generalization, this model has the potential to revolutionize various applications. However, its limitations, such as closed weights and US-only access, need to be addressed. As the model continues to evolve, it will be interesting to see how it is used in real-world scenarios and how it compares to other models in the market.