Is There a Meta Muse Startup Program?
No. Meta runs no dedicated startup program and issues no credits for Muse, Muse Spark, or the Meta Model API. You pay published rates from the first token.
Meta's developer programs page lists two offerings, both VR and MR only, and neither of them touches Muse, Muse Spark or the Meta Model API. Nor has Meta published any separate credit scheme covering them elsewhere.
That absence is the single most useful fact for a founder pricing this out, because the budgeting assumption most teams carry over from other providers does not hold here. Plan for a real invoice from day one. AI Perks tracks $7.7M in credits across 194 companies, which is where the offsets for a Muse-based stack realistically come from.
What Meta Shipped in 2026
Muse is not one product. It is six releases from Meta Superintelligence Labs, shipped across five months of 2026 under chief AI officer Alexandr Wang.
| Date (2026) | Release |
|---|---|
| April 8 | Muse Spark, the foundation model |
| July 7 | Muse Image |
| July 9 | Muse Spark 1.1 plus Meta Model API public preview |
| July | Muse Video previewed |
| August 5 | Muse Code, a coding agent aimed at large codebases |
| September 8 | Muse, the flagship consumer personal AI agent |
The ordering matters when deciding where to build. The developer surface arrived on 9 July, two full months before the consumer agent, and it is self-serve and compatible with the OpenAI SDK. In practice that usually means pointing an existing client at a different base URL and key rather than rewriting it.
The consumer agent that landed on 8 September is a different animal. It proactively completes tasks such as booking, filling forms, sending email and making purchases, running on a Meta-cloud virtual machine with a visible browser you can watch. Its permissions layer is called Sentinel. It ships on iOS, Android, WhatsApp and the web, with AI glasses listed as coming soon.
What It Costs to Build on Muse Spark
Muse Spark runs $1.25 per million input tokens and $4.25 per million output tokens on the Meta Model API. Muse Image is $0.01 per image, and Muse Voice Transcribe is $3.00 per 1,000 minutes.
| Meta Model API item | Published rate |
|---|---|
| Muse Spark 1.1, 1.2, 1.3 input | $1.25 per million tokens |
| Muse Spark 1.1, 1.2, 1.3 output | $4.25 per million tokens |
| Contributor tier input | $0.10 per million tokens |
| Contributor tier output | $0.20 per million tokens |
| Muse Image | $0.01 per image |
| Muse Voice Transcribe | $3.00 per 1,000 minutes |
Two things to read off that table. Output costs 3.4 times what input costs, so your bill is driven by how much the model writes, not how much you feed it. Sending the model a lot of input is comparatively cheap here; verbose generation is not. Trimming a chatty system prompt saves less than capping response length.
Second, no free credit grant is attached to any of it. Several third-party blogs claim new accounts receive a starting balance of free Meta AI credits. No such grant appears on Meta's own pricing page, so treat it as unconfirmed rather than budgeting against it.
The consumer app prices separately: a free tier, Power at $20 per month, and Maximum at $100 per month. Unusually, the tiers are differentiated by usage volume rather than by features, so a cheaper plan is a smaller allowance of the same product. Meta has not published the specific task or token ceilings for any of the three.
The Contributor Tier Trade: Cheaper Tokens for Your Data
The Contributor tier drops Muse Spark to $0.10 per million input and $0.20 per million output tokens, a 12.5x cut on input and a cut of just over 21x on output, in exchange for Meta training on your prompts and completions.
This is the closest thing to free Meta AI credits that verifiably exists, and it is not free. It is a data trade, priced honestly and disclosed up front, which is more than most discount tiers manage.
It fits a narrow but real set of workloads:
- Good fit: prototypes, internal tooling, evaluation harnesses, anything operating on public or synthetic data
- Bad fit: customer data, regulated inputs, or prompts that are themselves the product
That last one catches people. If your system prompt encodes the thing competitors would pay to know, a 12.5x discount is an expensive way to publish it. Route the cheap tier at the calls that are commodity work and keep the standard tier for anything carrying customer data or proprietary prompting. Meta prices both tiers openly, so the split is a deliberate choice rather than a default you drift into.
Why muse.ai Confuses Everyone
The domain muse.ai belonged to an independent video hosting and search company founded by Antonio Roldao. It rebranded to Skiv on 3 March 2026, and Meta subsequently took over the vacated domain as the web front end for its agent.
The rebrand set an embed URL migration deadline of 10 March 2026. The consequence for anyone researching now is that tutorials, forum answers and embed snippets written before March point at a domain that today serves a completely different company's product.
A third Muse adds to the mess. Microsoft announced a model called Muse in February 2025, a gameplay-generation research model known as WHAM, built with Xbox Game Studios and Ninja Theory. It has no connection to Meta whatsoever.
So before trusting any documentation, check which Muse it describes: Meta's agent and model family, Skiv's video platform, or Microsoft's game research model.
Where Meta Muse Credits Actually Come From
Since Meta issues none, the practical play is funding every other layer of the stack and paying Meta's rates out of what you saved.
Meta has run a Llama startup program for teams building on Llama models. Whether that program extends to Muse Spark or the Meta Model API is not confirmed by any public source, in either direction, so plan as though it does not until Meta says otherwise.
What is confirmed is that the layers around your model calls carry their own credit programs: cloud and GPU compute, vector databases, observability, analytics, payments, email and the rest of the operating stack. Whether any two of them combine depends on each program's own terms, so check before assuming they stack. Either way, the budget they free up is the budget a Muse Spark bill consumes. AI Perks covers 194 companies and $7.7M in total credits, refreshed as terms change.
Frequently Asked Questions
Does Meta offer free credits for Muse Spark?
No. Meta publishes no startup credit program for Muse Spark or the Meta Model API, and its developer programs page lists only VR and MR offerings. Claims of a free starting balance appear in third-party blogs but not on Meta's own pricing page. Credits for the surrounding stack are tracked at getaiperks.com.
How much does the Meta Model API cost?
Muse Spark 1.1, 1.2 and 1.3 cost $1.25 per million input tokens and $4.25 per million output tokens. Muse Image is $0.01 per image and Muse Voice Transcribe is $3.00 per 1,000 minutes. A Contributor tier cuts this to $0.10 and $0.20 if Meta may train on your data.
Is the Llama startup program the same as Meta AI credits?
They are not the same thing. Meta has run a startup program for teams building on Llama models, but no public source confirms whether it covers Muse Spark or the Meta Model API. Treat Muse as unfunded until Meta states otherwise, and cover the rest through AI Perks.
Is muse.ai the same as Meta Muse?
Today, yes, but it was not always. The muse.ai domain belonged to an independent video company that rebranded to Skiv on 3 March 2026, after which Meta took over the domain for its agent. Anything written about muse.ai before March 2026 describes the video platform, not Meta.
How much does the Muse app cost per month?
The consumer Muse agent has a free tier, a Power plan at $20 per month and a Maximum plan at $100 per month. The tiers differ by usage volume rather than by feature set. Meta has not published the specific task or token limits attached to any of the three plans.
What credits can I get instead of Meta AI credits?
Cloud compute, GPU, database, observability and developer tooling vendors run their own credit programs, with terms that vary by company and that decide whether any two of them combine. Meta funds none of your model calls, but the layers underneath them can be funded elsewhere. AI Perks tracks $7.7M across 194 companies at getaiperks.com.
Meta will not fund your Muse bill. Something else will, at getaiperks.com.