Muse Spark 1.1: Meta's New Agentic AI Model, API, and Pricing Explained

Muse Spark 1.1 is Meta's updated multimodal reasoning model, built specifically for agentic tasks: tool use, computer use, coding, and long-running multi-step work. It follows the original Muse Spark, which launched July 9, 2026, as the first model out of Meta Superintelligence Labs and the foundation of Zuckerberg's "personal superintelligence" push.

Meta just gave developers something they didn't have before: real access to its frontier model. On July 9, 2026, Meta Superintelligence Labs (MSL) released Muse Spark 1.1, an upgrade to the Muse Spark AI model it launched in April, and opened it to the public through a brand-new Meta Model API. Here's what changed, what it can do, and what it costs.

Where the original Muse Spark was aimed at consumers through meta.ai and the Meta AI app, 1.1 is Meta's first real move into the developer and enterprise market. For the first time, outside teams can build on Meta's frontier model instead of just using it inside Meta's own apps.

What's New in Muse Spark 1.1

A public API, finally. Meta launched the Meta Model API in public preview alongside 1.1, giving developers direct programmatic access. Previously, Muse Spark was locked to meta.ai and a small enterprise preview. New developers get $20 in free credits to start.

Pricing. The API is priced at $1.25 per million input tokens and $4.25 per million output tokens, undercutting Gemini 3.1 Pro ($2 / $12), GPT-5.4 ($2.50 / $15), and Claude Opus 4.6 ($5 / $25) by a wide margin.

A full 1 million token context window. The original Muse Spark topped out around 262K tokens. Muse Spark 1.1 manages a full 1M token window and can actively compact it, remembering earlier actions and retrieving relevant context from much earlier in a session while dropping what's no longer needed.

Agent orchestration, refined. Muse Spark 1.1 can act as either a main agent or a subagent. As the main agent, it gathers context, builds a plan, and delegates execution to parallel subagents. As a subagent, it stays within scope, knows which tools it has, and escalates back to the main agent when a task exceeds its authority. This is a more structured version of the "Contemplating mode" agent orchestration introduced in the original Muse Spark 1.1 ai model How to use.

Computer use. The model can now operate across multiple applications in a single session, adapting as information changes, keeping track of context over extended workflows, and navigating unfamiliar interfaces without much hand-holding.

Coding, substantially improved. Meta says Muse Spark 1.1 shows real gains on large, complex codebases: diagnosing and fixing bugs, shipping new features in enterprise-grade systems, and executing large-scale code migrations. This is Meta's direct answer to Claude and GPT models that currently dominate AI coding assistants.

Broader multimodal input. Muse Spark 1.1 now takes images, video, audio, and PDFs, on top of text. It's particularly strong at visual-to-code generation (turning a screenshot or mockup into working code) and detailed image and video captioning.

Meta just gave developers something they didn't have before: real access to its frontier model. On July 9, 2026, Meta Superintelligence Labs (MSL) released Muse Spark 1.1, an upgrade to the Muse Spark AI model it launched in April, and opened it to the public through a brand-new Meta Model API. Here's what changed, what it can do, and what it costs.

Where the original Muse Spark was aimed at consumers through meta.ai and the Meta AI app, 1.1 is Meta's first real move into the developer and enterprise market. For the first time, outside teams can build on Meta's frontier model instead of just using it inside Meta's own apps.

What's New in Muse Spark 1.1

A public API, finally. Meta launched the Meta Model API in public preview alongside 1.1, giving developers direct programmatic access. Previously, Muse Spark was locked to meta.ai and a small enterprise preview. New developers get $20 in free credits to start.

Pricing. The API is priced at $1.25 per million input tokens and $4.25 per million output tokens, undercutting Gemini 3.1 Pro ($2 / $12), GPT-5.4 ($2.50 / $15), and Claude Opus 4.6 ($5 / $25) by a wide margin.

A full 1 million token context window. The original Muse Spark topped out around 262K tokens. Muse Spark 1.1 manages a full 1M token window and can actively compact it, remembering earlier actions and retrieving relevant context from much earlier in a session while dropping what's no longer needed.

Agent orchestration, refined. Muse Spark 1.1 can act as either a main agent or a subagent. As the main agent, it gathers context, builds a plan, and delegates execution to parallel subagents. As a subagent, it stays within scope, knows which tools it has, and escalates back to the main agent when a task exceeds its authority. This is a more structured version of the "Contemplating mode" agent orchestration introduced in the original Muse Spark 1.1 ai model How to use.

Computer use. The model can now operate across multiple applications in a single session, adapting as information changes, keeping track of context over extended workflows, and navigating unfamiliar interfaces without much hand-holding.

Coding, substantially improved. Meta says Muse Spark 1.1 shows real gains on large, complex codebases: diagnosing and fixing bugs, shipping new features in enterprise-grade systems, and executing large-scale code migrations. This is Meta's direct answer to Claude and GPT models that currently dominate AI coding assistants.

Broader multimodal input. Muse Spark 1.1 now takes images, video, audio, and PDFs, on top of text. It's particularly strong at visual-to-code generation (turning a screenshot or mockup into working code) and detailed image and video captioning.

The April launch of Muse Spark was about proving Meta could compete on reasoning and multimodal understanding after the credibility hit from Llama 4's inflated benchmarks. It leaned into health reasoning (built with more than 1,000 physicians) and chart/image understanding, and it stayed consumer-only, with no public API and no downloadable weights, a sharp break from the open Llama tradition.

Muse Spark 1.1 keeps that reasoning core but retargets it at developers and agentic workflows: coding, tool use, computer use, and long-context tasks. It's less about answering your questions and more about doing multi-step work on your behalf, inside a codebase, a browser, or a stack of applications. Meta Muse Spark AI Tool: Complete Guide, Features, Benefits

Muse Spark 1.1 vs. the Competition

Model	API Pricing (input / output per 1M tokens)	Context Window	Public API
Muse Spark 1.1	$1.25 / $4.25	1M	Yes (public preview, July 2026)
Gemini 3.1 Pro	$2.00 / $12.00	1M	Yes
GPT-5.4	$2.50 / $15.00	1.05M	Yes
Claude Opus 4.6	$5.00 / $25.00	1M	Yes

On price alone, Muse Spark 1.1 undercuts every major frontier model currently on the market. Whether it holds up against GPT-5.4 and Claude Opus 4.6 on real coding and agentic benchmarks is still being tested independently, but the pricing move alone puts pressure on the rest of the field.

Should Developers Try It?

If you're building agentic tools, coding assistants, or computer-use workflows and want a cheaper option to test against your existing GPT or Claude integration, Muse Spark 1.1 is worth a look, especially with $20 in free API credits to start. If you need a proven track record in production or plan to fine-tune or self-host, the lack of open weights and the model's short time in public preview are worth weighing first.

The April launch of Muse Spark was about proving Meta could compete on reasoning and multimodal understanding after the credibility hit from Llama 4's inflated benchmarks. It leaned into health reasoning (built with more than 1,000 physicians) and chart/image understanding, and it stayed consumer-only, with no public API and no downloadable weights, a sharp break from the open Llama tradition.

Muse Spark 1.1 keeps that reasoning core but retargets it at developers and agentic workflows: coding, tool use, computer use, and long-context tasks. It's less about answering your questions and more about doing multi-step work on your behalf, inside a codebase, a browser, or a stack of applications. Meta Muse Spark AI Tool: Complete Guide, Features, Benefits

Muse Spark 1.1 vs. the Competition

Model	API Pricing (input / output per 1M tokens)	Context Window	Public API
Muse Spark 1.1	$1.25 / $4.25	1M	Yes (public preview, July 2026)
Gemini 3.1 Pro	$2.00 / $12.00	1M	Yes
GPT-5.4	$2.50 / $15.00	1.05M	Yes
Claude Opus 4.6	$5.00 / $25.00	1M	Yes

On price alone, Muse Spark 1.1 undercuts every major frontier model currently on the market. Whether it holds up against GPT-5.4 and Claude Opus 4.6 on real coding and agentic benchmarks is still being tested independently, but the pricing move alone puts pressure on the rest of the field.

Should Developers Try It?

If you're building agentic tools, coding assistants, or computer-use workflows and want a cheaper option to test against your existing GPT or Claude integration, Muse Spark 1.1 is worth a look, especially with $20 in free API credits to start. If you need a proven track record in production or plan to fine-tune or self-host, the lack of open weights and the model's short time in public preview are worth weighing first.

What is Muse Spark AI?

Muse Spark is a multimodal reasoning AI model from Meta Superintelligence Labs, first launched April 8, 2026, and built for text, image, and speech understanding with step-by-step reasoning.

What is Muse Spark 1.1?

Muse Spark 1.1 is the July 9, 2026 upgrade to Muse Spark, retargeted at developers with a public Meta Model API, a 1M token context window, computer use, and stronger coding and agentic performance.

Is Muse Spark 1.1 free to use?

The consumer version stays free at meta.ai and the Meta AI app. The developer API is paid, priced at $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits for new accounts.

How do I access the Muse Spark 1.1 API?

Sign up for the Meta Model API public preview, which is now open to developers. See our full walkthrough on using Muse Spark 1.1 for setup steps.

Is Muse Spark open-source?

No. Unlike Llama, Muse Spark and Muse Spark 1.1 are cloud-only, with no downloadable weights. Meta has said it "hopes" to open-source future Muse models but has given no timeline.

How does Muse Spark 1.1 compare to GPT-5.4 and Claude Opus 4.6?

Muse Spark 1.1 undercuts both on API pricing and matches their 1M token context window, but independent benchmarks still place it behind GPT-5.4 and Claude Opus 4.6 on coding and agentic tasks. See our complete guide to Muse Spark AI features and benefits for a full feature breakdown.