Meta releases Muse Spark Open open-weight multimodal model
The 816B total, 42B active parameter multimodal model features a 1M token context window and is optimized for both Apple silicon and standard inference engines.
- Supports native multimodal inputs across text, image, video, and audio for software engineering and multi-step reasoning.
- Runs on two Apple M5 Ultra Mac Studios via MLX, as well as on SGLang, vLLM, TensorRT-LLM, llama.cpp, and Unsloth.
- Available immediately via APIs on Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI.
- Weights are downloadable on Hugging Face, SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models under the Muse Model License.
Developers can run and customize a large-scale multimodal reasoning model locally on desktop Apple hardware or via major open-source inference stacks.
