What the Meta Model API Is, and Where Muse Spark API Fits
Meta Model API is Meta's developer platform, and Muse Spark 1.2 landed on it the day Muse Code shipped.
Meta Model API is the developer surface Meta serves its own models through, documented at dev.meta.ai. Muse Spark 1.2 arrived on it on 5 August 2026, the day Muse Code launched, so API and agent access were never separate release tracks; Meta's model listing is canonical.
Two identifiers matter. The default listing is muse-spark-1.2; muse-spark-1.2-contributor is the same model on cheaper terms, requiring you to grant Meta training rights over prompts and completions. That is a data decision before a billing one — our privacy breakdown reads both sets of terms, and the Muse Code pricing page holds the rates.
Muse Spark API gives you a model, not an agent. Background agents, the append-only event log, worktree parallelism and bundled skills belong to Muse Code, covered in our Muse Code explainer. Calling this API gets the model and none of that machinery.
How to Get Muse Spark API Access
Four steps take you from no account to a working first Muse Spark API call.
1. Create a Meta developer account
Sign up at dev.meta.ai with a Meta account, or use Login with Email if you would rather not link a social profile. You land in a developer console with nothing enabled yet. Community reports describe a selfie verification step, unconfirmed as universal but the most common reason people abandon sign-up.
2. Add a payment method
Attach a card first: no free allowance is advertised on Muse Spark API, and metering starts with token one. Community reports suggest a $20 starter credit existed during Muse Spark 1.1; whether it survives is unverified, so do not plan around it.
3. Generate a Muse Spark API key
Create a key in the console and store it in an environment variable rather than a source file — one credential any OpenAI-style client can carry.
export MUSE_API_KEY="your-key-here"4. Make a first call and pick a tier
Send a short completion request using model id
muse-spark-1.2, ormuse-spark-1.2-contributoronce you accept training terms. A successful response confirms key and tier together; watch response headers for your rate-limit budget, since tiers differ fiftyfold on requests per minute.
Muse Spark API Rate Limits by Tier
Rate limits differ sharply between tiers, and the cheaper tier is also restricted by geography.
| Tier | Requests / min | Tokens / min | Data terms | Availability |
|---|---|---|---|---|
| Standard (muse-spark-1.2) | 3,000 RPM | 4M TPM | Meta states it does not train on your data | Broad |
| Contributor (muse-spark-1.2-contributor) | 60 RPM | 2.1M TPM | You grant Meta training rights on prompts and completions | Some countries only |
Limits as reported, cross-checked against Meta's pricing and rate limits documentation. Prices live on the Muse Code pricing page, with a Muse Code cost calculator for your own volume.
Using the Muse Spark API Through OpenRouter
OpenRouter lists meta/muse-spark-1.2, and its per-key spend limits solve a real Muse Code billing problem.
OpenRouter carries Muse Spark as meta/muse-spark-1.2, the shortest path if you already route other models through it: one billing relationship, one key format, a client most agent frameworks speak.
A second reason is less obvious. Meta's billing has no hard spending cap, only email alerts — the loudest complaint from early Muse Code users, summed up on Hacker News as being one mistake away from a large bill. OpenRouter supports per-key spend limits, so routing Muse Spark API traffic through it caps an uncapped account. That workaround is community practice, not an official recommendation, but it is the cleanest fix available.
Trade-off: a middleman, meaning extra margin on token prices and one more service between you and Muse Spark. For small, bursty usage the cap earns it.
Muse Spark Playground: Is a Card Required?
A browser playground exists, though the card requirement around it is community-reported, not documented.
Meta's developer console includes a browser playground for trying Muse Spark 1.2 without writing client code — the fastest way to sanity-check prompts before wiring them into an application.
Whether the playground demands a payment method is where reporting gets thin. Community reports suggest a card must be on file before it will run a request, consistent with no advertised free tier, but Meta has not documented the rule, so treat it as unverified. Playground usage is metered like any other Muse Spark API call.
Running Muse Spark in opencode, Cline and Other Harnesses
Muse Spark API is model-only, so third-party harnesses are how people get agent behaviour outside Muse Code.
Because Muse Spark sits behind a standard API, it drops into harnesses that let you bring your own key. Developers have wired Muse Spark 1.2 into opencode and Cline, generally through OpenRouter — also the route into IDE tools accepting custom model endpoints.
People searching for a "muse spark cli" want one of two things. Meta's own agent is Muse Code — install it with our Muse Code install guide, and read the troubleshooting page when sign-in misbehaves. For Muse Spark inside a CLI you already trust, bring your key to opencode, Cline or Aider.
One asymmetry: Muse Code is bound to Muse Spark 1.2 with no published model-selection option, so substitution runs one way only. Our Muse Code alternatives page compares those harnesses.
What Is Not Available in the Muse Spark API
Three gaps come up constantly in Muse Code searches, and all three answer the same way: not yet.
- No AWS Bedrock listing. Muse Spark 1.2 is not offered through Bedrock, so procurement running through AWS Marketplace has no path today.
- No open weights. There is no checkpoint to self-host and no Hugging Face repository. Zuckerberg said he would have more to share on that soon, as reported by The Register — our Muse Spark 1.2 page tracks that question.
- No free tier, no hard cap. Billing starts at token one, and Meta provides alerts rather than a ceiling.
None of this is hidden — it is what a platform one day past launch has not built yet. Re-check all three against this page's verification date.
Related Muse Spark and Muse Code Guides
Where the prices, specs and agent behaviour behind this Muse Spark API guide are covered.
Frequently Asked Questions
Is the Muse Spark API free?
No. No free tier is advertised on Muse Spark API, and a payment method is required before your first call. Community reports suggest a $20 starter credit existed during Muse Spark 1.1, but whether it survived into 1.2 is unverified. See the Muse Code pricing page for rates.
How do I get a Muse Spark API key?
Create a developer account at dev.meta.ai, add a payment method, then generate a key in the console and keep it in an environment variable. The sequence takes minutes unless identity verification stalls you, which several early Muse Code users reported.
What are the Muse Spark API rate limits?
Standard allows 3,000 requests and 4M tokens per minute; Contributor drops to 60 requests and 2.1M tokens per minute and is limited to certain countries. That fiftyfold gap on requests matters more than the token gap for parallel Muse Code workloads. Both figures are as reported — verify in Meta's documentation.
Is Muse Spark available on AWS Bedrock?
No. Muse Spark 1.2 has no Bedrock listing, so AWS-native procurement is not an option. Two routes exist: Meta Model API and OpenRouter, which lists meta/muse-spark-1.2. Nothing published suggests a Bedrock deal is coming, so plan around direct billing.
Can I set a spending limit on the Muse Spark API?
Not on Meta's side. The platform offers email alerts but no hard ceiling, the most common billing complaint from early Muse Code users. The community workaround is OpenRouter's per-key spend limits, or a virtual card with its own cap. Both are practices, not official features.
Do I need a credit card for the Muse Spark playground?
Community reports suggest yes, and Meta has documented no rule either way. With no free tier advertised anywhere on the platform, assuming a card is required is the safe planning position for Muse Code users. Playground calls are metered like any other request.
Is there a Muse Spark CLI separate from Muse Code?
No official one. Muse Code is Meta's terminal client for Muse Spark, and no standalone Muse Spark CLI has been published. To use it in a different command-line tool, bring your key to opencode, Cline or Aider, usually via OpenRouter. See our install guide for Meta's agent.
Can I use the Muse Spark API without installing Muse Code?
Yes, and many people do. Muse Spark API is independent of the agent: call the model from any client without installing Muse Code, which matters because Muse Code runs on macOS and Linux only. The reverse is not true — Muse Code is bound to Muse Spark 1.2 with no documented model swap.
Sources
- Meta developer docs — pricing and rate limitsdev.meta.ai
- OpenRouter — meta/muse-spark-1.2openrouter.ai
- Meta developer — Muse Spark model listingdeveloper.meta.com
- Meta AI Research — Introducing Muse Code and Muse Spark 1.2research.meta.ai
- Hacker News — billing and rate-limit discussionnews.ycombinator.com