Together AI adds Muse Spark Open to serverless and dedicated infrastructure

Together AI made Meta’s Muse Spark Open available on October 15 through its serverless API and dedicated infrastructure. This gives developers two managed ways to use the open-weight model without deploying it themselves.

Standard serverless rates are $1.25 per 1 million input tokens, $4.25 per 1 million output tokens and $0.15 per 1 million cached input tokens, according to Together AI’s published pricing. The same page displayed a limited promotion that set input and output charges at zero through October 18 at 12 p.m. PT. That promotion is separate from the standard rates.

Those per-token figures apply to the serverless service. The opened materials do not list a Muse Spark Open-specific price, hardware configuration, capacity commitment or provisioning process for dedicated infrastructure.

Together AI identifies the serverless endpoint as meta-models/Muse-Spark-Open-816B-A42B and provides sample calls through its chat completions API. Together AI’s model page says the managed endpoint accepts text, image, audio and video inputs and returns text.

The company advertises a one-million-token context window. Meta’s launch announcement and Meta’s model card specify the same limit. Meta describes Muse Spark Open as a mixture-of-experts model with 816 billion total parameters and 42 billion active parameters.