Meta Launches Muse Spark 1.1: First Paid Frontier Model API

Meta Superintelligence Labs has officially released Muse Spark 1.1, a new multimodal reasoning model specifically engineered for agentic tasks. Alongside this release, Meta has launched a public preview of the Meta Model API, marking a significant strategic shift for the company. While Meta has historically distributed its models as open weights, Muse Spark 1.1 is a closed, hosted, and metered service, representing the company’s first paid API for a frontier model.

Capabilities and Agentic Performance

Muse Spark 1.1 is designed as a reasoning model, meaning it processes information and plans before generating responses. The model supports a 1,048,576-token context window and accepts inputs including text, images, video, and documents. A core feature of the model is its ability to manage this large context window through active compaction, allowing it to remember actions and retrieve information from long sessions.
The model is built for orchestration, excelling in tool use, computer use, and multimodal understanding. It can function as a primary agent that gathers context and delegates tasks to parallel subagents, or as a subagent that executes specific jobs. Meta reports that the model demonstrates zero-shot generalization to new native tools, custom skills, and Model Context Protocol (MCP) servers. In computer use scenarios, the model is trained to either write scripts for automation or perform direct interactions like clicking, depending on which method is more efficient.

Benchmarks and Integration

According to Meta’s internal benchmark evaluations, Muse Spark 1.1 leads in tool-use and tool-augmented reasoning tasks. While it ranks behind competitors such as Opus 4.8 and GPT-5.5 in coding-specific benchmarks like SWE-Bench Pro and DeepSWE 1.1, it is positioned as a highly capable orchestration model. The model supports structured output, parallel tool calling, and prompt caching, with a web_search tool available for generating cited answers.
Integrating the model into existing workflows is designed to be straightforward, as the Meta Model API is OpenAI-compatible. Developers can migrate by updating the base URL in their existing setups. The API also supports Anthropic-format harnesses, allowing for easy adoption in various agentic environments.

Pricing and Availability

The public preview of the Meta Model API is currently available in the United States, with no access provided for the European Union at this time. Developers are charged based on usage, with pricing set at $1.25 per million input tokens and $4.25 per million output tokens. New accounts are eligible for $20 in free credits. For individual consumers, the model is available for free in Thinking mode through the Meta AI app and the meta.ai website.