Muse Spark and Meta Muse are not the same thing. Muse Spark is an AI model, announced by Meta Superintelligence Labs on April 8, 2026. Meta Muse is a personal agent app, launched on September 8, 2026, that uses Muse Spark as its brain. Think of Muse Spark as the engine and Muse as the car built around it, with its own computer, security system and price list.

What is Muse Spark?

Meta introduced Muse Spark on April 8, 2026 as the first model in Meta Superintelligence Labs' Muse series (Meta). Meta described it as:

  • "small and fast by design";
  • offering Instant and Thinking modes;
  • able to run parallel subagents.

It is not open-weight. Meta offered a private preview API and said it hopes to open-source future versions. By the time the agent launched, Meta was calling Muse Spark "Meta's most capable model to date" (Meta Newsroom).

Muse's security post names Muse Spark 1.3 and calls it "close to SOTA" on prompt-injection resistance (Muse blog). Third-party analysis from eesel says version 1.3 trails Anthropic's Opus 5 on several agent benchmarks and is verbose, but fast and cheap.

What is Meta Muse?

Meta Muse is the consumer product. Launched in the US on September 8, 2026 under the internal code name "Hatch", it is a personal agent that sends email, books travel, fills forms, shops and negotiates on your behalf, available on iOS, Android, the web and WhatsApp.

Muse adds a lot around the model:

  • a dedicated Linux VM per user with its own browser;
  • Sentinel, a separate security service outside the agent's container that allows, denies or asks about every connector action and network request;
  • surrogate tokens, so the agent never sees real passwords;
  • connectors for email, calendars, payments, shopping and more;
  • plans: a free tier, Power at $20/month and Maximum at $100/month.

Our Meta Muse review covers the product in depth, and Sentinel explained covers the security layer.

Muse vs Muse Spark side by side

Muse Spark Meta Muse
What it is AI model Personal agent app
Announced April 8, 2026 September 8, 2026
Who uses it Developers via private preview API; powers Meta products Consumers on iOS, Android, web, WhatsApp
Version named 1.3 (Muse security post) Single app, updated continuously
Price (as of November 2026) No public consumer price Free tier; Power $20/month; Maximum $100/month
Can act on websites Not on its own Yes, via its VM and browser
Open source No (future versions possibly) No

Why does the naming matter?

The distinction affects how you read news and benchmarks:

  • Benchmark results for Muse Spark tell you about raw model ability, not how well the agent handles CAPTCHAs, approvals or shopping sites.
  • Security incidents involving Muse, such as the September macOS zero-day or the Amazon block, concern the app and its infrastructure, not the model.
  • Pricing applies to the Muse app. Muse's plans meter "Muse tokens" (500M a week on Power, 3B on Maximum), which are a product allowance, not the API price of the model. See Meta Muse pricing.

Where does the confusion come from?

Part of it is timing. Meta announced the Muse Spark model in April, months before any consumer could use the Muse app, so early coverage used "Muse" as shorthand for the model. When the app launched in September, headlines used the same word for the product. Search results still mix the two, and some third-party pages describe Muse Spark features, such as Thinking mode, as if they were app settings.

Other names that cause confusion

Muse Code

Meta also sells Muse Code, a separate developer agent with $5, $15 and $50 plans (Meta for Developers). It is aimed at programmers and is not the consumer Muse agent.

Muse for small business

In late September 2026 Meta extended Muse to small businesses. That is the same Muse product line applied to business accounts, not a different model.

Gemini Spark

Google's Gemini Spark is an entirely different product: Google's always-on personal agent, announced at I/O on May 19, 2026 and built on Gemini 3.5. The shared word "Spark" is a coincidence. If you meant Google's agent, read our Gemini Spark review.

Muse Charm

Meta has also mentioned a Muse Charm keychain device as a coming way to reach the agent. It is hardware for Muse, not a model.

Does the model choice matter for Muse users?

For most people, only indirectly. Muse has no model picker, so you cannot swap Muse Spark for another model inside the app. What the model affects is speed, cost and how well the agent resists manipulation. Meta chose a model it describes as small and fast, which helps explain why Muse can offer a free tier and a $20 Power plan while OpenAI's dots start at $100 a month on a larger flagship model. Meta also says Muse Spark 1.3 is close to the state of the art on prompt-injection resistance, but it pairs the model with Sentinel because no model alone is trusted to police itself. Our Meta Muse vs OpenAI dots comparison covers how those choices play out in practice.

Quick way to tell them apart

  1. Is it something you download or chat withthat is Meta Muse, the agent
  2. Is it in a benchmark table or API docsthat is Muse Spark, the model
  3. Is it for writing code with $5 to $50 plansthat is Muse Code
  4. Is it from Google with Gmail and Chromethat is Gemini Spark

For definitions of related terms such as harness, Sentinel and SOUL.md, see the personal AI agent glossary. UK readers waiting for the app can follow personalagent.uk's Muse launch tracker.