Muse Spark Open
An open-source Mixture-of-Experts large language model with 816B parameters, supporting NVFP4 4-bit quantization, suited to distributed inference deployment.
Core Facts
Muse Spark Open can be called on the day of release through platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
Muse Spark Open uses an MoE architecture, with 816B total parameters and 42B activated parameters.
Muse Spark can be used as a backend via OpenCode Go.
Timeline (last 90 days)
The announcement did not disclose Muse Spark Open's parameter scale, benchmark results, or license details.
Meta's claim of being "America's strongest open-weight model" currently lacks third-party evaluation support.
Muse Spark Open enters the open-weight model segment competing with models such as the Llama series and Mistral.
As an inference service provider, Together AI is responsible for deploying Muse Spark Open to production, so developers can call it without building their own GPU clusters.
In Unsloth's UD-Q4_K_XL GGUF quantized versions, Muse Spark Open is about 506 GB and Qwen3.8-Flash-Next is about 111 GB.
Meta's 816B-parameter model Muse Spark Open scores 39 on the Artificial Analysis benchmark.
Muse Spark Open requires submitting legal name, date of birth, affiliated organization and other information, requires prominently displaying "Built with Muse Spark Open" in the user interface, and requires a separate commercial license application for annual revenue exceeding $20 million.
Meta released two models, Muse Wonder and Muse Spark Open; Muse Wonder focuses on agents and professional work, and Muse Spark Open is an 816B-parameter MoE open-weights model.
The version run via Ollama's cloud is not entirely the same as the open-weights version in terms of resource scheduling—the cloud version removes the hardware barrier of local deployment, but the real value of open weights lies in being able to download the weights and self-host them.
Muse Spark Open can be run in the cloud on Ollama with the command `ollama run muse-spark-open:cloud`.
40 more timeline events
All Facts (20)
Muse Spark Open can be called on the day of release through platforms such as Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
90%VerifiedMuse Spark Open uses an MoE architecture, with 816B total parameters and 42B activated parameters.
85%VerifiedMuse Spark can be used as a backend via OpenCode Go.
85%VerifiedMuse Spark is a model specifically designed for coding assistants and intelligent agents.
75%VerifiedMeta released the open weights of Muse Spark Open on October 15, calling it officially the strongest US open-weight model to date.
85%VerifiedMeta publicly stated on August 10 that it would release the weights of the Muse Spark model as open source.
65%Partially VerifiedThe 816B-parameter Muse Spark Open occupies about 493GB of memory after NVFP4 quantization, while FP16 storage would require more than 1.6TB.
80%Partially VerifiedThe default models of multiple Providers have changed: Cortecs, CrossModel, Eden AI, Google Gemini, GitHub Copilot, Google Vertex AI, LLM Gateway, Ofox, OpenCode Go, and Prime Inference now default to Gemini 4 Argon; Fireworks AI, Nebius Token Factory, and Requesty switch to Muse Spark Open; CoreWeave, DigitalOcean, NanoGPT, and Nvidia adopt Nemotron 3.5 Super VL; Aixy and Neon use Claude Haiku 5.5; Agnes AI, Ambient, IteraCompute, Poe, Scaleway, and SenseNova also have their own adjustments.
75%Partially VerifiedMuse Spark Open has a parameter scale of 816B, with 42B activated parameters.
75%Partially VerifiedMuse Wonder supports a 1 million Token context window.
75%UnverifiedMuse Spark Open runs on vLLM and targets real production deployment scenarios.
70%UnverifiedThe actual retrieval accuracy of Muse Spark Open under a 1M context window has not yet been verified and requires developers to test it themselves.
70%Partially VerifiedMuse Spark Open uses the Muse Model License.
70%Partially VerifiedMuse Spark Open is positioned as the most powerful open-weight model in the United States to date.
70%Partially VerifiedMuse Spark Open supports a 1 million token context window.
70%Partially VerifiedThe Muse model license requires prominent labeling of 'Built with Muse Spark Open', and companies with annual revenue over $20 million that offer it as a service must notify Meta and sign a commercial agreement.
65%Partially VerifiedHugging Face calculates that the KV cache under BF16 is about 250 KB per token, and a full 1-million-token context requires about 266 GB.
65%Partially VerifiedMuse Spark Open has 816 billion total parameters, activates 42 billion parameters per token, and has a context window of 1 million tokens.
65%Partially VerifiedMuse Spark Open's weight download channels include Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models.
65%Partially VerifiedMuse Spark Open can run on two Mac Studios equipped with M5 Ultra chips, together with the latest MLX framework.
65%