Meta Superintelligence Labs has released Muse Spark Open, the open-weight version of its Muse Spark model. Muse Spark Open is an 816B-parameter Mixture-of-Experts model that activates 42B parameters per token. It takes text, image, video and audio input, returns text, and holds a 1M-token context window. Meta team claims it to be the most capable US open-weight model to date. For developers, the real news is deployment: the NVFP4 build runs on 2 Apple M5 Ultra Mac Studios.
TL;DR
- Size: 816B total / 42B active parameters (256 experts, 8 active per token); 1M-token context.
- Runs on: 2 Mac Studios (M5 Ultra, 512 GB each) via MLX; 4 Blackwell Ultra GPUs or 8 H200s with NVFP4; about 300 GB setups via Unsloth 2-bit GGUF.
- Performance: Strong on agentic tool use; mid-pack among large open models on coding.
- Best: MCP-Atlas 89.2, the top score in Meta’s comparison table.
- Worst: Terminal-Bench 2.1 76.4, behind Kimi K3 (85.0), GLM 5.3 (83.9) and Qwen3.8 (82.0).
What is Muse Spark Open?
Muse Spark Open is a natively multimodal MoE model from Meta, built for coding and long-horizon agentic work. Weights ship in BF16 and NVFP4 on Hugging Face. The open version received extra safety post-training. Meta team states that this makes it more robust without system-level guardrails.
How does the architecture work?
The model card lists 62 layers and a hidden dimension of 8192. Each MoE layer holds 256 experts, and top-k routing activates 8 per token. Attention is grouped-query, with 128 query heads and 16 KV heads. A separate 1.8B-parameter Perception Encoder handles visual input. The vocabulary has 202,048 tokens, and training data covers more than 100 languages.
Reasoning depth is set with reasoning_effort: low, medium, high or xhigh. The default is xhigh.
How does Muse Spark Open perform on benchmarks?
Meta compares it with Kimi K3, GLM 5.3, Qwen3.8-2.4T-A95B and others at maximum reasoning effort. Its strongest results are on tool use and knowledge reliability:
- MCP-Atlas: 89.2, ahead of Qwen3.8 (84.5) and GLM 5.3 (84.2).
- JobBench: 61.8, narrowly ahead of GLM 5.3 (61.4).
- HLE-Diamond, no tools: 26.1, versus 22.2 for Kimi K3.
- AA Omniscience: 25, the top score in Meta’s table.
- ZeroBench with tools: 50.0, versus 41.0 for Kimi K3.
Coding is the weak spot. On Terminal-Bench 2.1 with the Terminus 2 harness, it scores 76.4. Kimi K3 reaches 85.0, GLM 5.3 scores 83.9 and Qwen3.8 scores 82.0. Its Artificial Analysis Intelligence Index is 39. That sits behind GLM 5.3 (45), Kimi K3 (44) and Qwen3.8 (40). So “most capable US” holds, but Chinese open models still lead overall.
Some rival scores are self-reported; the evaluation methodology lists each source.