How to Use Muse Spark 1.1: Meta's New Agentic AI Model (Complete Guide)

Learn how to use Meta's Muse Spark 1.1 AI model: free access on meta.ai, the paid API, pricing, key features, and real use cases. Full guide + FAQ. Meta released Muse Spark 1.1 on July 9, 2026, its second model from Meta Superintelligence Labs and a major upgrade over the original Muse Spark. It is a multimodal reasoning model built for agentic tasks, meaning it does not just answer questions, it plans, uses tools, operates computers, writes code, and manages multi-step projects on its own. If you want to know what Muse Spark 1.1 actually does and how to start using it today, this guide covers everything from free consumer access to the paid developer API.

Muse Spark 1.1 is a closed-weight, proprietary AI model built for agentic computing. Unlike Meta's open-weight Llama family, it cannot be downloaded or fine-tuned locally. It sits at the top of Meta's model lineup and is designed to compete directly with GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro.

Its standout features are a 1 million token context window, the ability to orchestrate multiple subagents in parallel, strong computer-use skills across apps, and gains in coding and multimodal reasoning (text, image, and audio input, though output is currently text-only).

Muse Spark 1.1 is a closed-weight, proprietary AI model built for agentic computing. Unlike Meta's open-weight Llama family, it cannot be downloaded or fine-tuned locally. It sits at the top of Meta's model lineup and is designed to compete directly with GPT-5.5, Claude Opus 4.8, and Gemini 3.1 Pro.

Its standout features are a 1 million token context window, the ability to orchestrate multiple subagents in parallel, strong computer-use skills across apps, and gains in coding and multimodal reasoning (text, image, and audio input, though output is currently text-only).

How to Access Muse Spark 1.1

There are two ways to use the model, depending on whether you are a casual user or a developer.

1. Free access through the Meta AI app and meta.ai

Muse Spark 1.1 runs in "Thinking" mode inside the Meta AI app and on meta.ai. To use it:

Open the Meta AI app or go to meta.ai and log in with a Meta account.
Start a new chat and switch the mode to "Thinking." This activates Muse Spark 1.1 instead of the lighter default model.
Type your task in plain language. Since the model is agentic, phrase requests as goals rather than single questions, for example "research three competitor pricing pages and summarize the differences" instead of a one-line factual question.
Attach images or audio if your task needs visual or audio context. The model accepts these as input, though it will only reply in text.
Let it work through multi-step tasks. For longer projects, Muse Spark 1.1 will plan, delegate parts of the work to subagents, and keep track of context across the session.

This consumer access is free but likely rate-limited during heavy use.

2. Paid access through the Meta Model API

Developers can build with Muse Spark 1.1 through the new Meta Model API, which launched in public preview alongside the model.

Join the Meta Model API waitlist for developer access (public preview, US developers first).
Once approved, generate an API key from the developer console.
Use the API the same way you would call OpenAI's API, since Meta built it to be OpenAI-compatible. This means existing OpenAI SDK code often needs only the base URL and API key changed.
Send requests with your prompt, any tool definitions (native tools, MCP servers, or custom skills), and your desired output format. Muse Spark 1.1 supports structured output and parallel tool calling.
Monitor usage against pricing: $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits before you're billed pay-as-you-go.

Because the API ships OpenAI-compatible, most existing agent frameworks (LangChain, OpenCode, Cline, and similar tools) can point at Muse Spark 1.1 with minimal changes to your integration.

What You Can Actually Do With It

Multi-agent project orchestration. Give it a broad goal like "pull last quarter's support tickets, categorize them, and draft a summary deck," and it will split retrieval, tagging, and drafting across parallel subagents rather than working sequentially.

Cross-app computer use. It can operate a browser or desktop across multiple applications when information changes mid-task. Meta's own demo has it placing a dinner order and adjusting it mid-session when new information comes in.

Agentic coding. It can diagnose bugs in large codebases, implement new features, and run large code migrations. It supports planning mode, goal conditioning, subagent delegation, and context compaction, and works inside coding harnesses like OpenCode, Cline, and Replit.

Multimodal-to-action workflows. It can watch a smartphone video, extract product photos, reason about the item, and then operate a browser to create a marketplace listing, combining perception and action in one pass, not unlike the groundwork covered in our guide on running Facebook ads in the UAE.

Tips for Getting Better Results

Give it outcome-based goals rather than micromanaged steps, since it performs best when allowed to plan its own subtasks. Provide relevant tools, MCP servers, or custom skills upfront if your workflow needs them; Meta says the model zero-shot generalizes to new tools without extra fine-tuning. For long sessions, don't worry about repeating earlier context, the model actively manages its 1M-token window and compacts older steps on its own. If you need image or video generation, pair Muse Spark 1.1 with Meta's separate Muse Image model, since Spark's own output is text-only.

Pricing Summary

Access type	Cost
Meta AI app / meta.ai (Thinking mode)	Free, with likely rate limits
Meta Model API	$1.25 per million input tokens, $4.25 per million output tokens
API free tier	$20 in free credits, then pay-as-you-go

How to Access Muse Spark 1.1

There are two ways to use the model, depending on whether you are a casual user or a developer.

1. Free access through the Meta AI app and meta.ai

Muse Spark 1.1 runs in "Thinking" mode inside the Meta AI app and on meta.ai. To use it:

Open the Meta AI app or go to meta.ai and log in with a Meta account.
Start a new chat and switch the mode to "Thinking." This activates Muse Spark 1.1 instead of the lighter default model.
Type your task in plain language. Since the model is agentic, phrase requests as goals rather than single questions, for example "research three competitor pricing pages and summarize the differences" instead of a one-line factual question.
Attach images or audio if your task needs visual or audio context. The model accepts these as input, though it will only reply in text.
Let it work through multi-step tasks. For longer projects, Muse Spark 1.1 will plan, delegate parts of the work to subagents, and keep track of context across the session.

This consumer access is free but likely rate-limited during heavy use.

2. Paid access through the Meta Model API

Developers can build with Muse Spark 1.1 through the new Meta Model API, which launched in public preview alongside the model.

Join the Meta Model API waitlist for developer access (public preview, US developers first).
Once approved, generate an API key from the developer console.
Use the API the same way you would call OpenAI's API, since Meta built it to be OpenAI-compatible. This means existing OpenAI SDK code often needs only the base URL and API key changed.
Send requests with your prompt, any tool definitions (native tools, MCP servers, or custom skills), and your desired output format. Muse Spark 1.1 supports structured output and parallel tool calling.
Monitor usage against pricing: $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits before you're billed pay-as-you-go.

Because the API ships OpenAI-compatible, most existing agent frameworks (LangChain, OpenCode, Cline, and similar tools) can point at Muse Spark 1.1 with minimal changes to your integration.

What You Can Actually Do With It

Multi-agent project orchestration. Give it a broad goal like "pull last quarter's support tickets, categorize them, and draft a summary deck," and it will split retrieval, tagging, and drafting across parallel subagents rather than working sequentially.

Cross-app computer use. It can operate a browser or desktop across multiple applications when information changes mid-task. Meta's own demo has it placing a dinner order and adjusting it mid-session when new information comes in.

Agentic coding. It can diagnose bugs in large codebases, implement new features, and run large code migrations. It supports planning mode, goal conditioning, subagent delegation, and context compaction, and works inside coding harnesses like OpenCode, Cline, and Replit.

Multimodal-to-action workflows. It can watch a smartphone video, extract product photos, reason about the item, and then operate a browser to create a marketplace listing, combining perception and action in one pass, not unlike the groundwork covered in our guide on running Facebook ads in the UAE.

Tips for Getting Better Results

Give it outcome-based goals rather than micromanaged steps, since it performs best when allowed to plan its own subtasks. Provide relevant tools, MCP servers, or custom skills upfront if your workflow needs them; Meta says the model zero-shot generalizes to new tools without extra fine-tuning. For long sessions, don't worry about repeating earlier context, the model actively manages its 1M-token window and compacts older steps on its own. If you need image or video generation, pair Muse Spark 1.1 with Meta's separate Muse Image model, since Spark's own output is text-only.

Pricing Summary

Access type	Cost
Meta AI app / meta.ai (Thinking mode)	Free, with likely rate limits
Meta Model API	$1.25 per million input tokens, $4.25 per million output tokens
API free tier	$20 in free credits, then pay-as-you-go

Agentic AI models like Muse Spark 1.1 are becoming a practical production tool for agencies running high-volume campaigns, not just a novelty. A team producing client videos can use its multimodal reasoning to caption and tag raw footage before an editor opens the file, the same groundwork we cover in our guide to working with an AI video production agency in Dubai. Agencies managing paid social can point it at a browser to draft, review, and adjust ad copy in real time, a workflow that pairs well with the fundamentals in our online advertising agency in Dubai guide.

Because it operates across multiple apps in one session, it also fits naturally into the reporting and account-management work that eats up hours at any Google and Facebook ads agency in Dubai. And since agentic tools shift how much manual work a campaign needs, it is worth revisiting your numbers against current social media marketing costs in Abu Dhabi, since lower execution time changes what a fair monthly retainer looks like.

Limitations to Know Before You Start

Muse Spark 1.1 is closed-weight, so there is no local deployment or custom fine-tuning option, unlike Llama. Independent benchmarks show it trails Claude Opus 4.8 and GPT-5.5 on the hardest long-horizon coding tasks (DeepSWE 1.1), even though it leads on agentic tool-use benchmarks like MCP Atlas and JobBench. Documentation on the API is still sparse, with no detailed model card at launch.

FAQ

What is Muse Spark 1.1? It is Meta's second Superintelligence Labs model, released July 9, 2026, a multimodal reasoning model built for agentic tasks like tool use, computer use, and coding, with a 1 million token context window.

Is Muse Spark 1.1 free to use? Yes, for consumers. It's free in Thinking mode on the Meta AI app and meta.ai with a Meta login. Developer access through the Meta Model API is paid, with $20 in free credits to start.

How much does the Muse Spark 1.1 API cost? $1.25 per million input tokens and $4.25 per million output tokens.

Is Muse Spark 1.1 open source like Llama? No. It is proprietary and closed-weight. There is no local deployment or community fine-tuning, unlike Meta's Llama models.

Can Muse Spark 1.1 generate images or video? No. It accepts text, image, and audio as input but only outputs text. For image generation, use Meta's separate Muse Image model.

How is Muse Spark 1.1 different from the original Muse Spark? 1.1 adds major gains in tool use, computer use, coding, and multi-agent orchestration, plus active management of its 1M-token context window, making it faster on complex, multi-step projects.

How does Muse Spark 1.1 compare to GPT-5.5 and Claude Opus 4.8? It leads on agentic tool-use benchmarks like MCP Atlas and JobBench, but trails both on the hardest long-horizon coding benchmark, DeepSWE 1.1.

Is the Meta Model API compatible with existing tools? Yes. It's built to be OpenAI-compatible, so most existing OpenAI SDK integrations and agent frameworks need minimal changes to work with it.