Meta releases Muse Spark Open, touted as highest‑performing open‑weights model in US
Meta announced on Oct. 15, 2026 that it is making a new AI model called Muse Spark Open available for download. ITmedia AI+ reported that the model is being touted as the highest‑performing open‑weights system produced in the United States. The weights are posted on Hugging Face, so anyone can download them and run the model locally on their own hardware.
Key points
- Muse Spark Open has 816 billion parameters and uses MoE, activating about 42 billion per query.
- Weights are publicly released on Hugging Face, enabling local execution on high‑end hardware.
- Running the model requires two Mac Studios with M5 Ultra, achieving ~25 tokens/sec per pair.
Muse Spark Open contains 816 billion parameters and uses a Mixture‑of‑Experts (MoE) architecture that activates roughly 42 billion parameters for each query. Reports say the model can read text, images, video and audio but only generates text output. To run it at usable speed, Apple’s verification indicated that two Mac Studio machines equipped with M5 Ultra chips are needed, delivering about 25 tokens per second per pair of units (around 36 tokens per second with four units). Commercial use is permitted under certain conditions, but the model is far beyond the capability of a typical laptop.
The release highlights a shift toward open‑weight, locally runnable AI, which may appeal to organizations that must keep data on‑premise or want to avoid cloud‑based inference costs.
The story so far
3 episodes →- Meta releases Muse Spark Open, touted as highest‑performing open‑weights model in USthis story
[Meta×AI] What is Muse Spark Open? | For those who only use ChatGPT, a 3-minute guide to AI that runs without the internet
note.com · 17 October 2026
Loading the full article…
This text was published by note.com and written by AI活用ラボ. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Google unveils Gemini 4 Argon with 1 million-token output · 1 src
- DeepSeek releases DeepSeek-V4.1-Flash-1017, scores 77.0 on DeepSWE v1.1 and opens Search and Fetch APIs · 6 src
- Kyutai releases REKI 2.4B vision-language-action model for robot manipulation · 1 src
- NVIDIA releases Nemotron 3.5 Super VL and Bolt edge models for enterprise agents · 11 src
- Ggml-org merges CUDA patch to accelerate low-bit MMVQ in llama.cpp · 1 src
Comments
via GitHub Discussions