# Meta Muse models

> Use Meta’s standard Muse Spark and Muse Wonder models with credits or your own key.

Use Meta’s Muse Spark models through the same Clonvo Chat Completions endpoint as other providers. No separate SDK is required.

## Models

| Clonvo model id | Input / output / cached input per 1M tokens | With data sharing |
| --- | --- | --- |
| `meta/muse-spark-1.3` | $1.25 / $4.25 / $0.15 | $0.10 / $0.20 / $0.002 |
| `meta/muse-spark-1.2` | $1.25 / $4.25 / $0.15 | $0.10 / $0.20 / $0.002 |
| `meta/muse-spark-1.1` | $1.25 / $4.25 / $0.15 | Not available |

The [model catalog](https://clonvo.com/models?provider=meta) lists current capabilities and prices. Muse Wonder 1.0 was retired on October 31, 2026; requests for it return `404 model_retired`. Review [Meta’s data commitments](https://dev.meta.ai/legal/commitments) for its processing and retention practices.

## Data-sharing prices

Muse Spark 1.3 and 1.2 have a data-sharing version (Meta’s Contributor mode) at a much lower price, in exchange for Meta being allowed to train on those requests. Clonvo uses it automatically for organizations that have turned on **Use free models** in **Settings → Privacy**, whose terms include data sharing. You keep using the same model id, and you’re charged the data-sharing price. Clonvo doesn’t store these requests.

-   To use the standard version for one request, send `"provider": { "data_collection": "deny" }`.
-   The data-sharing version has much lower rate limits. When it’s busy, Clonvo serves the request from the standard version at the standard price.
-   It doesn’t offer the `max` reasoning effort, so `max` runs as `xhigh` there.

A request’s details in **Activity** show which version served it: the upstream model is `muse-spark-1.3-contributor` for data sharing.

## Make a request

```sh
curl https://api.clonvo.com/v1/chat/completions \
  -H "Authorization: Bearer $CLONVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta/muse-spark-1.3",
    "messages": [{"role": "user", "content": "Explain HTTP caching briefly."}],
    "reasoning_effort": "minimal",
    "max_completion_tokens": 1024
  }'
```

Text, image input, streaming, function tools and JSON output (`json_schema` and `json_object`) work through Chat Completions. Meta chooses tools automatically: `tool_choice: "none"` sends the request without tools, and `"required"` or a named function runs as `"auto"`, so the model may answer without calling a tool. Image URLs are downloaded by Meta, so they must be publicly reachable; otherwise send the image as a base64 data URL.

Reasoning tokens are billed as output tokens, even though Meta does not expose its private reasoning text. They consume the completion-token budget, so leave enough room for reasoning and the visible answer.

Muse always reasons. Use `minimal`, `low`, `medium`, `high` or `xhigh`; Muse Spark 1.3 also accepts `max`, and other efforts are matched to the nearest supported one. Reasoning can’t be turned off: `reasoning_effort: "none"` or `reasoning.enabled: false` runs at `minimal`.

Clonvo removes unsupported stop sequences, presence/frequency penalties, log probabilities and verbosity settings rather than forwarding them to Meta. Native image generation, transcription, PDF/audio input, search grounding and Meta’s Responses API are not exposed by this integration. See [Meta’s API guide](https://dev.meta.ai/docs/protocols/chat-completions).

## Credits or your own key

For Clonvo credits, use the usual paid-model flow. To use your own Meta account, create a key in the [Meta developer dashboard](https://dev.meta.ai), then add it under **Provider keys → Meta**. Your own Meta key needs the Pay as you go plan or higher, and the usual [BYOK fees and fallback settings](https://clonvo.com/docs/byok.md) apply.