Muse Spark Open

Meta's open-weight 816B-parameter MoE model (42B active, 1M context) for coding and long-horizon agent work, released Oct 15, 2026.

Meta · Open-weight/community models · model

Published in Feed

Overview

Muse Spark Open is Meta's first open-weight release of its Muse Spark line, announced October 15, 2026. It is a natively multimodal mixture-of-experts model with 816B total and 42B active parameters and a 1M-token context window, aimed at coding, multi-step reasoning, and long-horizon agentic tasks. Weights are published on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models under the Muse Model License, which carries limited commercial terms. Meta says it runs on two Apple M5 Ultra Mac Studios using MLX.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • 816B total / 42B active parameter mixture-of-experts architecture
  • 1M-token context window, natively multimodal
  • Open weights on Hugging Face, SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models
  • Hosted API access via Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI
  • Runs on two Apple M5 Ultra Mac Studios using MLX
  • Supports SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, and llama.cpp
  • Custom Helion GPU kernels in SGLang with 1.03-3.95x kernel-level speedups on NVIDIA and AMD GPUs

Use cases

  • Self-hosted coding and agent backends where data cannot leave the organization
  • Long-context document and codebase reasoning
  • Local inference on high-memory Apple silicon workstations
  • Fine-tuning a large open-weight base under the Muse Model License

Pricing

Weights are free to download under the Muse Model License (limited commercial terms; exact thresholds not confirmed). Hosted API pricing varies by provider and is not stated by Meta.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
15 Oct 2026
Record updated
16 Oct 2026
Record last checked
16 Oct 2026

Explore the Feed

Other entries filed under open-weight/community models; inclusion does not imply equivalent features.

  • Nemotron 3.5 Bolt

    NVIDIA's open 74B MoE model (9B active) packed into 24GB for local agents on RTX Spark; released Oct 15, 2026.

  • Jais 3

    Six open Arabic-focused language models from 900M to 375B parameters with 512K context, plus Jais Speech for Arabic recognition.

  • kettei

    Sqwish Labs' decision-model family returning answer probabilities in one pass, with open 3.7B and 900M models and a fine-tuning API (OneHot).

Browse the Feed