What Is Meta Muse?
Meta Muse is a personal AI agent that Meta released on September 8, 2026, and also the family name covering every model Meta Superintelligence Labs shipped during 2026. The consumer agent is the flagship. Underneath it sit a foundation model, an image model, a video model and a coding agent, all carrying the same Muse brand.
Meta Superintelligence Labs, the division run by chief AI officer Alexandr Wang, built the entire line in roughly five months. The agent runs on iOS, Android, WhatsApp and the web at muse.ai, with AI glasses listed as coming soon.
The useful question for a founder is not what Muse does in a demo. It is what it costs, what Meta has actually published, and whether there is a credit program behind it. On that last point the answer is no, and AI Perks tracks $7.7M in credits across 194 companies that do run one.
Every Muse Release in 2026
Meta shipped six Muse releases between April and September 2026, going from foundation model to consumer agent in five months.
| Date (2026) | Release | What it is |
|---|---|---|
| April 8 | Muse Spark | The foundation model the family is built on |
| July 7 | Muse Image | Image generation |
| July 9 | Muse Spark 1.1 + Meta Model API | Self-serve public preview, OpenAI SDK compatible |
| July | Muse Video | Previewed, video generation |
| August 5 | Muse Code | Coding agent aimed at large codebases |
| September 8 | Muse | The flagship consumer personal AI agent |
Two entries matter more than the rest. The Meta Model API went self-serve on July 9 and is OpenAI SDK compatible, which means moving an existing integration over is a base URL and key swap rather than a rewrite. And Muse Code, on August 5, was positioned for large codebases rather than for line completion, which puts it in a different bracket from autocomplete tools.
What the Muse Agent Actually Does
Muse is proactive: it completes tasks instead of describing how to complete them. Meta positions it around booking, filling in forms, sending emails and making purchases on your behalf.
The architecture is the interesting part. Each session runs on a Meta-cloud virtual machine with a visible browser, so you watch the agent click through the same web pages a person would. The steps are there to be watched, which is not true of an agent that acts through a headless API call.
Permissions are handled by a layer Meta calls Sentinel. That is the component deciding what the agent is allowed to do without asking you first, which is the whole trust question for any agent that can spend money.
What Meta has not published is equally worth knowing. There is no stated task ceiling or token allowance for the free tier. Several third-party write-ups quote numbers, but Meta's own materials do not, so treat any specific free-tier limit you see as unverified.
What Meta Muse Costs
The consumer app has three tiers - free, Power at $20 per month and Maximum at $100 per month - and they are separated by usage volume rather than by features.
| Tier | Price | What you get |
|---|---|---|
| Free | $0 | The agent, at the lowest usage volume. Ceiling not published |
| Power | $20 / month | Higher usage volume |
| Maximum | $100 / month | Highest usage volume |
That structure is worth reading carefully. A features-based ladder tells you which capabilities are locked. A volume-based ladder tells you the product is expected to be used until it runs out, which is how agents behave once they work.
Developers pay separately through the Meta Model API. These are Meta's own published developer prices:
| Meta Model API product | Price |
|---|---|
| Muse Spark 1.1 / 1.2 / 1.3 | $1.25 per million input tokens, $4.25 per million output tokens |
| Contributor tier (same models) | $0.10 per million input tokens, $0.20 per million output tokens |
| Muse Image | $0.01 per image |
| Muse Voice Transcribe | $3.00 per 1,000 minutes |
The Contributor tier is the line to stop on. It is roughly 92% cheaper on input and 95% cheaper on output, and the price of that is consent: Meta trains on your prompts and completions. On a workload of 50M input and 10M output tokens a month, standard pricing costs $105 and Contributor costs $7.
That is a real decision, not a pricing gimmick. For an internal tool or a public content pipeline, the discount is close to free money. For anything touching customer data, proprietary prompts or unreleased product copy, it is not a tier you can use, and the standard rate is what you budget. Credits on the surrounding stack are tracked on AI Perks, which is where the savings have to come from instead.
Muse, muse.ai and Microsoft Muse Are Three Different Things
The name is shared by three unrelated products, and most search results for "muse ai" still conflate them.
muse.ai was an independent video hosting and search company founded by Antonio Roldao. It rebranded to Skiv on 3 March 2026, with an embed URL migration deadline of 10 March. Meta subsequently took over the vacated muse.ai domain as the web front end for its agent. If you have older bookmarks, tutorials or embed codes referencing muse.ai video hosting, those belong to Skiv, not to Meta.
Microsoft Muse, announced in February 2025, is a gameplay generation research model also known as WHAM, built with Xbox Game Studios and Ninja Theory. It has no connection to Meta whatsoever.
So when a search result promises "Muse AI pricing", check which Muse it means before you plan a budget around it.
Are There Meta Muse Credits for Startups?
No. Meta runs no dedicated Muse startup program and no Muse credit program. Its developer programs page lists only Meta Horizon Start and the Meta Horizon Creator Program, both of which are VR and MR only and unrelated to Muse.
Meta does operate a Llama startup program, but whether it extends to Muse Spark is unconfirmed by any source either way. And the "$20 in free Meta Model API credits for new accounts" figure circulating on third-party blogs does not appear on Meta's official pricing page. Budget as though neither exists.
That absence is the practical takeaway. Muse is a list-price line item for now, which makes it more important, not less, to cover the rest of your stack with credits that genuinely exist. AI Perks indexes the programs that do, by category.
| Stack layer | What credit programs on AI Perks typically cover |
|---|---|
| Frontier model APIs | Hosted inference on large language models |
| GPU and serverless compute | Training runs, inference hosting |
| Cloud infrastructure | Databases, storage, egress |
| Developer and GTM tools | Observability, analytics, CRM, email |
Total tracked on getaiperks.com: $7.7M in credits across 194 companies.
The useful way to read that index is by your own invoice. Filter to the categories that match what you actually pay for each month rather than the ones with the biggest headline numbers, because a large credit on a service you do not use is worth nothing and a small one on your largest line is worth its face value.
Then treat Muse as the exception rather than the rule. If Muse Spark or the consumer agent sits in your stack, that part is a cash cost with no program behind it, which is exactly the argument for making sure no other layer is paying list price at the same time.
Frequently Asked Questions
What is Meta Muse?
Meta Muse is a personal AI agent released on September 8, 2026 by Meta Superintelligence Labs, and the brand name for Meta's 2026 model family: Muse Spark, Muse Image, Muse Video and Muse Code. The agent runs on iOS, Android, WhatsApp and the web, with AI glasses listed as coming soon.
How much does Meta Muse cost per month?
The consumer app has three tiers: free, Power at $20 per month and Maximum at $100 per month. They differ by usage volume rather than by features, and Meta has not published the free tier's task ceiling. Developer access is billed separately through the Meta Model API.
Is muse.ai the same company as Meta Muse?
No. muse.ai was an independent video hosting and search company founded by Antonio Roldao that rebranded to Skiv on 3 March 2026. Meta later took over the vacated domain as the web front end for its agent. Microsoft's Muse, a gameplay research model, is a third unrelated product.
What is Sentinel in Meta Muse?
Sentinel is the permissions layer governing what the Muse agent may do on your behalf. Because Muse runs on a Meta-cloud virtual machine with a visible browser and can book, buy and send email, Sentinel is the control deciding which of those actions proceed without asking you first.
Does Meta give free credits for Muse?
No. Meta's developer programs page lists only Meta Horizon Start and the Meta Horizon Creator Program, both VR and MR only. No Muse credit program exists, and the "$20 free API credits" claim on third-party blogs is unverified. Credits from 194 other companies are tracked at getaiperks.com.
What is the Contributor tier in the Meta Model API?
Contributor prices Muse Spark at $0.10 per million input tokens and $0.20 per million output tokens instead of $1.25 and $4.25, in exchange for Meta training on your prompts and completions. It suits internal tooling and public content, and it rules itself out for anything involving customer data. Offset the rest via AI Perks.
Meta runs no Muse credit program. 194 other companies do run one.