- Published on
A Multimodal Reasoning Model for Agentic Tasks on Meta Model API
- Authors
- Name
- DeepHorus
Meta Superintelligence Labs released Muse Spark 1.1 today, a multimodal reasoning model for agentic tasks, alongside a public preview of the Meta Model API. The model is closed, hosted, and metered per token, marking a strategic shift from Meta's previous open-weight distribution. This change matters because developers now gain access to a reasoning model explicitly built for orchestration in complex, multi-step workflows.
Muse Spark 1.1 builds on its predecessor with reported gains in tool use, computer use, coding, and multimodal understanding. It features a context window of 1,048,576 tokens, adjustable reasoning effort per request, and support for inputs including text, images, video, and documents. The API also offers structured output, parallel tool calling, a Files API, and prompt caching.
Pricing is set at $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. In performance benchmarks published by Meta, Muse Spark 1.1 leads in tool-use and tool-augmented reasoning but ranks third in coding and multimodal tasks.
The model is designed as an orchestration engine rather than a coding-accuracy leader. It actively manages its million-token context through compaction, remembering and retrieving earlier information efficiently. Delegation allows it to act as a main agent gathering context and planning, or as a subagent following instructions and escalating when needed.
Computer use incorporates the same logic, generating scripts when automation is faster and clicking when direct interaction is simpler. The API is OpenAI-compatible, so migration requires only a base URL change. This compatibility, along with support for Anthropic-format harnesses, lowers the barrier for integrating Muse Spark 1.1 into existing stacks.
Developers may find Muse Spark 1.1 valuable for tasks like multimodal listing automation, screenshot-driven debugging, and adaptive planning. The closed API model ensures consistent performance but limits accessibility due to regional restrictions. Early reactions indicate interest in its tool-use capabilities, though some note the lack of EU availability as a drawback.
Looking ahead, Meta is likely to expand regional access and add more features. The emphasis on orchestration and tool use positions Muse Spark 1.1 as a strong contender in the agentic AI space, especially for teams needing reliable, scalable reasoning models.
