‹ All stories

Muse Spark Open

Product · 3 updates

Where it stands

As of Oct 16, NVIDIA published Muse-Spark-Open-816B-A42B-DFlash2 on Hugging Face, a 4.38B-parameter draft checkpoint that speeds up serving of Meta's Muse-Spark-Open-816B-A42B through DFlash 2 speculative decoding with its Model Optimizer.

Timeline · newest first
  1. Official

    Infron adds Meta's open-weight Muse Spark Open model to its platform

    @InfronAI on X published this on Oct 16, 2026 (source: x.com).

    Infron said on October 16 that Muse Spark Open is now available on Infron. It describes the model as a natively multimodal open-weight model with 816B total parameters that takes text, image, audio and video input and is built for coding, tool use and long-horizon agents.

    Source: X @InfronAI
  2. Official

    NVIDIA releases a DFlash 2 speculative-decoding draft model for Meta's Muse Spark Open 816B

    NVIDIA published this on Oct 16, 2026 (source: huggingface.co).

    NVIDIA published Muse-Spark-Open-816B-A42B-DFlash2 on Hugging Face, a 4.38B-parameter draft checkpoint that speeds up serving of Meta's Muse-Spark-Open-816B-A42B through DFlash 2 speculative decoding with its Model Optimizer. The drafter has five dense decoder layers with sliding-window attention and shares the target model's token embedding and output head; DFlash 2 adds a two-tap in-block convolution around every sublayer and a pairwise path selector over each position's top-k candidates. It is released in BF16 under the NVIDIA Open Model Agreement for commercial and non-commercial use, supports TensorRT-LLM and vLLM on Blackwell GPUs, and was trained on prompts from Nemotron-Post-Training-Dataset-v2.

    Source: NVIDIA
  3. Official

    Meta releases Muse Spark Open, an 816B open-weights multimodal model with 42B active parameters and 1M-token context that runs on two M5 Ultra Mac Studios

    Meta published this on Oct 15, 2026 (source: research.meta.ai).

    Meta released the weights for Muse Spark Open, a natively multimodal mixture-of-experts model with 816B total and 42B active parameters and a 1M-token context window, built for coding and agentic work; Meta calls it the most capable US open-weight model to date. Meta says it is the first model in its class to run on a desktop using two Apple M5 Ultra Mac Studios with the latest MLX. Weights are on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models under the Muse Model License with limited commercial terms, and API access is live at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, which prices it at $1.25 per 1M input and $4.25 per 1M output tokens. Support covers SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp and Unsloth.

    Source: Meta
Related stories
Meta Muse6 updates · Apps & products→Meta2 updates · Apps & products→NVIDIA36 updates · Models→Claude33 updates · Models→
How we label
Official
An institution announces something that has already happened, through its own channel or a founder or executive account.
Statement by …
A named person's views, predictions, plans or comments, including remarks reported by media when the original is unavailable.
Reported by …
Only media or another third party is saying it. We name the outlet.
Unconfirmed
Circulating online (posts, comments, leaks). No company, named person or outlet has confirmed it.

Items older than 72 hours never go in Top stories and show their original date.

DailyCitedRSSAboutPrivacy