Meta unveils Muse Spark Open to compete with Chinese open-weights
The multimodal model has 816bn total parameters, with 42bn active parameters and a 1m-token context window. Read more: Meta unveils Muse Spark Open to compete with Chinese open-weights
Meta has introduced Muse Spark Open, positioning it as the most capable US open-weight model currently available. This multimodal model boasts an impressive 816bn total parameters, with 42bn active parameters and a generous 1m-token context window. Meta emphasizes that Muse Spark Open is the first in its class to run seamlessly on a desktop computer using Apple's M5 Ultra Mac Studios through their latest machine learning framework.
Designed for complex, multi-step reasoning tasks such as coding and agentic work, Muse Spark Open stands out from closed weight competitors. Unlike closed weight models, developers can freely download and modify the open weight model to suit their specific needs. These open weights are distinct from open source models that provide access to both training data and code.
The open-weight model can be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models. Muse Spark Open is also accessible via popular API platforms including Groq, OpenRouter, and Baseten. China leads the pack in the open-weights space, with models from Chinese developers like Xiaomi, Z AI, Kimi, Alibaba, and DeepSeek consistently outperforming US counterparts in the Artificial Analysis ranking.
To ensure safety, Muse Spark Open has undergone additional post-training safety evaluations, meeting the stringent thresholds set by Meta.
Written by urgent.news from Silicon Republic's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- "Mommy and daddy" play nice: Apple and Meta team up to bring open-source Muse to Macs axios.com
- Meta’s Muse Spark Open isn't quite so open economictimes.indiatimes.com
