Muse Spark Open: Hardware, License and Cost to Self-Host
Almost nobody should self-host Muse Spark Open: the model card asks for at least 700 GB of VRAM even in NVFP4, so for anything under billions of tokens a month the hosted endpoint at $1.25 per million input tokens is the cheaper and simpler route. Meta released the open-weight 816B-parameter mixture-of-experts model on 15 October 2026[1]. Only 42B parameters are active per token, which helps…
Self-hosting Meta's Muse Spark Open model is challenging due to its size and hardware requirements. The model consists of 816 billion total parameters, but only 42 billion are active per token. 1 trillion tokens of context are supported, with a 1% of the weights being used for each token. Running the model requires at least 700 GB of VRAM for NVFP4 and 2 TB for BF16, making it expensive to self-host.
Meta offers hosted endpoints at $1.25 per million input tokens and $4.25 per million output tokens. The model is multimodal, built for coding and agentic tasks, and supports token windows of up to 1 million. It is available for use, fine-tuning, distillation, and redistribution, but selling it as a Model-as-a-Service requires a separate agreement with Meta once gross revenue exceeds $20 million in a 12-month period.
The model's architecture includes 256 experts and 8 activated per token, with a hidden dimension of 8192 and a perception encoder of 1.8 billion parameters. The model is available for inference on various platforms such as SGLang, vLLM, Hugging Face Transformers, TensorRT-LLM, Unsloth, and llama.cpp.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
