Meta has released Muse Spark Open, an open-weight version of its Muse model, on Hugging Face. Meta's chief AI officer Alexandr Wang called it "the most capable US open-weight model to date," according to Khaleej Times. Open-weight means developers can download the model's parameters and run or adapt it themselves, unlike closed models such as ChatGPT or Claude.
What you get
- A mixture-of-experts model that reads text, images, video and audio and writes text, with a 1 million token context window.
- Trained on data in more than 100 languages, according to the model card.
- Support in common tools including vLLM, Hugging Face Transformers, llama.cpp, LM Studio and Ollama.
The fine print
The Muse Model License allows use, modification, fine-tuning and redistribution. Products built on it must say "Built with Muse Spark Open." Companies that sell access to the model as a service and earn more than $20 million a year must notify Meta and sign a separate agreement first; internal use and building it into your own product are excluded.
It is not a laptop model. The model card says the NVFP4 weights need at least 700GB of GPU memory and the full BF16 weights about 2TB. Meta says it is optimised to run on two Mac Studios with M5 Ultra chips.
How it compares
Meta's own table is candid. On the Artificial Analysis Intelligence Index it scores 39, against 40 for Qwen 3.8, 44 for Kimi K3 and 45 for GLM 5.3. It leads on some agent and tool-use tests, and on refusing harmful requests. Independent evaluator Vals ranks it 38th of 49 models on its index, with its best result, third of 79, on a finance agent test.
