Muse Spark Open

The checked Meta announcement lists hosting partners and downloadable weights. It does not publish a Meta-operated paid API tariff. Together publishes $1.25 input, $4.25 output, and $0.15 cached input per million tokens for its hosted service.

The checked sources do not establish a free Muse Spark Open API entitlement. Together’s published hosted tariff is $1.25 input, $4.25 output, and $0.15 cached input per million tokens. Check the exact endpoint and current account terms.

Muse Spark Open weights are available under the Muse Model License. Muse Wonder is a separate model, and its API or subscription price does not apply to Muse Spark Open.

No. Context capability is a limit description, while a token bill depends on usage categories and the hosting tariff. At Together’s published rates, one million uncached input tokens alone would have a $1.25 input component, but any billed output adds its own charge. One million eligible cached input tokens would have a $0.15 input component instead. Those arithmetic components do not establish that an account can send a request at the model’s maximum context length or that every token in that request is cache-eligible. Check the serving endpoint’s current limit and its billing documentation. A monthly total of one million tokens can also consist of many smaller requests, so an aggregate budget cannot by itself validate an individual prompt. Keep context validation and cost calculation as separate checks.

Begin with the exact model and route, then record input, output, and eligible cache-read prices in the provider’s original currency. Keep its context limit and source date alongside those rates. Apply the same assumed usage categories to each offer only when their billing categories are comparable. If another provider has no published cache price, do not assume its cache reads are free or substitute Together’s rate. For the Together reference, ten million uncached input tokens and two million output tokens produce a $21 token estimate; a different provider needs its own rates before the same calculation can be made. The estimate is not a latency test or a quality ranking. It also does not establish a Meta-operated API tariff, because the cited paid service is supplied by Together. Any account credit or temporary promotion should remain separately identified.