Meta Muse Pricing: Free, Power, Maximum and the Muse Spark API

Meta’s pricing is confusing mainly because two different things share one name. Muse Spark is the AI model. Muse is the personal agent app that runs on it, and it does tasks for you rather than only answering questions. If that split is new, what Muse Spark is and how it relates to the Muse app sets it out before you look at any prices.

So there are really two price lists. One is for people who want the agent to do things on their behalf. The other is for developers paying for the model by the token. This post covers both, in that order, then works through how to decide which one you need.

What to check before comparing plans

A monthly price on its own tells you very little about an agent. Four questions matter more:

• —How much use each tier allows. Agent plans are sold by usage, not by feature list, so the question is whether a tier covers your week.
• —What you have to hand over to start. A payment card, a social login or your prompts can all be part of the price.
• —Whether your data trains the model. On some tiers that is the trade that makes them cheaper.
• —Which product you actually need. The chatbot, the agent app and the developer API are billed separately.

Muse, the agent app: three tiers

As of September 2026, Muse comes in three tiers, and the figures below are from reporting at launch. Prices of this kind move, so check the app before you commit.

• —Muse Free, with a usage meter, according to TechCrunch’s launch report.
• —Power, at $20 a month, from the same TechCrunch report.
• —Maximum, at $100 a month, also reported by TechCrunch.

Is Meta Muse free? Partly. The free tier exists, but 9to5Mac reports that it has weekly usage limits. TechCrunch also reports that a payment card is required to start, even on the free tier. According to the same report, Meta expects most people to stay on free.

What does the Power plan get you over free? The published reporting describes the tiers by usage, not by a list of extra features, so the honest answer is more room on the meter. The same goes for Power vs Maximum. Nothing in the launch coverage says Maximum adds a capability Power lacks; it is the tier for people who hand over much more work.

All three tiers run the same model, Muse Spark 1.3, and connectors are switched on one at a time. The launch coverage lists email, calendars, payments, shopping, dining and more, with a browser used where no direct connection exists. Muse is rolling out in the US on iOS, Android and muse.ai, per Meta’s own announcement. The practical side of getting it running is in setting up the Meta Muse app.

The chatbot is a different product

Meta’s chatbot on meta.ai also runs on Muse Spark, and it is not the same thing as the Muse agent. Simon Willison’s hands-on write-up notes that meta.ai needs a Facebook or Instagram login. The three tiers above are the ones reported for the Muse app. If all you want is to ask questions, draft text or look something up, the chatbot is the place to start, and the agent tiers only become relevant when you want it to act for you.

Muse Spark API pricing for developers

Developers pay for the model directly through the Meta Model API. Standard pricing for Muse Spark 1.3 is $1.25 per million input tokens and $4.25 per million output tokens, according to Artificial Analysis. The same analysis lists cached input at $0.15 per million, which matters if your app sends the same long instructions or documents again and again.

There is also a cheaper contributor tier, and the price is paid in data. The Register reports that it requires letting Meta use your prompts for training. Meta’s Muse Code post adds that the contributor tier is rate-limited in a rolling five-hour window and available in select countries, with standard pay-as-you-go pricing as the alternative. Developers who only want the coding agent can also pick one of Muse Code’s monthly plans, which start at $5 a month. If you work with client code or anything confidential, the contributor tier is probably the wrong one, whatever it saves.

Per-token prices are only half the story, because models use different numbers of tokens to finish the same job. Artificial Analysis ran its intelligence index and measured the cost per task instead. On that measure, Muse Spark 1.3 at its xhigh setting cost $0.55 per task, against $0.58 for Gemini 3.8 Flash and $1.23 for Claude Opus 5 on its high setting. That is a useful signal on cost. It is still a test set, not your work, and a model that is cheap per task on a benchmark can take more attempts on your particular problem.

How it compares with Claude and Google

On the consumer side, the price points look familiar. Claude’s pricing page lists a free plan, Pro at $20 a month ($17 billed annually) and Max from $100 a month. Google’s Gemini subscriptions run from free through Google AI Plus at $4.99 and Google AI Pro at $19.99 to Google AI Ultra from $99.99 a month.

So Muse’s two paid steps sit close to what the other two charge. The difference is in what the money buys. Claude Pro adds Claude Code, Claude Cowork and Research on top of more usage. Google AI Pro includes Gemini Spark in select countries, Deep Research and the Jules coding agent. Muse’s tiers, as reported, are about how much the agent can do for you. A closer look at how the agents themselves differ is in Muse Spark vs Gemini.

On the API side, Claude’s pricing page lists Sonnet 5 at $2 input and $10 output per million tokens, and Opus 5 at $5 and $25. Against those, Muse Spark 1.3’s standard rate in the Artificial Analysis write-up is lower per token. Whether it is lower for your job depends on how many tokens each model needs and how often it gets the answer right first time.

Worked example: is free enough for your week?

Nobody can tell you in advance how far the free meter goes for you, because it depends on what you ask for. The practical method is to count the tasks you would really hand over, not the ones that sound impressive. Finding the repetitive part of your work is the same exercise applied to a job.

Take one ordinary week and write down every errand you would give an agent if it were free and reliable. For one person, it might look like this:

1. 01Monday: turn three saved recipe reels into a grocery list.
2. 02Tuesday: draft replies to two emails, for approval before sending.
3. 03Wednesday: fill in a school trip form.
4. 04Thursday: compare train times for a weekend visit and hold the best option.
5. 05Friday: send invitations for a birthday dinner and track who has replied.

That is five tasks, most of them short. Now sort them. A grocery list or two email drafts are small jobs. Anything that opens a browser, works through several pages or keeps running after you close the app is likely to use more of the meter. If your list is mostly small jobs, start on free and watch the meter for a full week before you decide anything.

The upgrade question then answers itself. If you hit the weekly limit by Wednesday on tasks you genuinely wanted done, Power is worth trialling. If you only reach it on a week with a trip to plan, stay on free and accept the odd wait. Maximum makes sense when the agent is doing substantial work every day, which for most people means it is part of their job. The kinds of work people reasonably give to an assistant are covered in using AI as an executive assistant.

One more filter before you pay for more usage. Some of the tasks on your list may be ones an agent should not do unsupervised, however cheap it is. What AI is actually bad at is worth a read before you hand over anything involving money or other people.

The caveat: free costs something

Three honest limits on everything above. First, tiers differ by usage, and the reported descriptions do not give you a precise allowance, so the only reliable test is your own week. Second, pricing in this market changes often, and these are launch figures.

Third, free is rarely free. The Muse free tier still asks for a payment card, the chatbot asks for a Facebook or Instagram login, and the cheaper developer tier is paid for with your prompts. Meta says you can opt out of your Muse interactions being used to train its models, and the setting is described in Meta’s Muse announcement. Decide on that before you connect your email, not after. What the app can see and how to limit it is covered in whether Meta Muse is safe.

The short version

Muse has a free tier with weekly limits, then Power at $20 and Maximum at $100 a month, and the difference between them is how much work the agent does. The Muse Spark API is priced per token, $1.25 in and $4.25 out per million, with a cheaper contributor tier that trades your prompts for the discount. Count your real tasks for a week before paying for anything.

The plan matters less than knowing what to give an agent and how to check what comes back. That is what Coursium teaches: using AI tools well at work, in short lessons on your phone. You can find out more on the Coursium home page.