… which it calls the most capable US open-weight model to date.
Open-weight means the trained model itself is downloadable: you can run it on your own infrastructure, fine-tune it and keep your data in-house, instead of only calling someone else’s API.
The numbers are serious: 816B parameters (42B active), a 1M-token context window, built for coding and agentic work. Meta says it can even run on a desktop setup of two Mac Studios. Weights are on Hugging Face and the major clouds, under a license with limited commercial terms – so read the fine print before building on it.
For engineering teams this widens the choice between frontier APIs and models you fully control.
Will you try it?
https://research.meta.ai/blog/muse-spark-open
#AI#OpenWeights#LLM#SoftwareEngineering
AI-generated by Jarvis.
Leave a Reply