Meta 发布开源权重模型 Muse Spark Open，总参数 816B、激活参数 42B，支持 1M token 上下文窗口，原生支持文本、图像、视频和音频多模态，面向编码与智能体任务。该模型在 NVIDIA GB300 NVL72 上训练，Meta 同时产出针对 Blackwell Ultra 优化的 NVFP4 checkpoint，单 GPU 吞吐超过 50K tokens/sec。NVIDIA 已在博客中介绍相关细节，NVFP4 版本发布在 Hugging Face。
Meta 开源 816B 权重模型并给出 NVFP4 量化版本，可了解大模型在 Blackwell 上的训练与推理吞吐表现。
And @AIatMeta went live with Muse Spark Open!
The 816B model was trained on NVIDIA GB300 NVL72. Its 1M-token context window gives coding agents room to work across large codebases.
Meta also produced NVFP4 checkpoints optimized for Blackwell Ultra, delivering over 50K tokens/sec/GPU to improve throughput and lower cost per token.
Congrats to the @Muse team.
More in our blog: https://nvda.ws/4B8d9g4
NVFP4: https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B-NVFP4
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date. Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Key Model Specs: → 816B total / 42B active parameters → 1M token context window → Natively multimodal across text, image, video, audio在 X 查看被引用的帖子
来源：NVIDIA AI · x.com