Meta open-sources Muse Spark Open: 816B params, 1M context, runs on two M5 Ultra Mac Studios

jiayi_pirate · x · 2026-10-16

Meta released open weights for Muse Spark Open, billed as the most capable US open-weight model to date.

Specs

• 816B total / 42B active params, natively multimodal, 1M-token context
• Built for coding and agentic work: long-horizon tasks, complex software engineering, multi-step reasoning

Desktop deployment

• First in its class to run at home on two Apple M5 Ultra Mac Studios (up to 512GB unified memory each) with the latest MLX framework

Availability

• APIs: Baseten, Fireworks, Groq, Modal, Nebius, OpenRouter, Together AI
• Weights: Hugging Face, SageMaker, Google Cloud Model Garden, Microsoft Foundry
• Engines: SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp, Unsloth; optimized for AMD/Apple/Intel/NVIDIA

Benchmarks (vs Inling, Nemotron 3 Ultra, GLM 5.3, Kimi K3)

• Leads MCP-Atlas (89.2), JobBench (61.8), HLE-Diamond (26.1), AA-Omniscience (25)
• Trails on GDPval-AA v2 (1401 vs GLM 5.3's 1653) and Terminal-Bench 2.1 (76.4 vs Kimi K3's 85.0); SciCode 58.5 near Kimi K3's 59.5
• Leads multimodal-with-tools: ZeroBench 50.0, BabyVision 80.2, SimpleVQA 73.6

Scale AI CEO Alexandr Wang promoted the release; ex-DeepSeek/ByteDance researcher Edward Sun said it matches the best model his team shipped two months ago — now open-weight.

Related event: Meta Open-Sources Muse Spark Open, an 816B-Parameter Model It Calls America's Best(40 posts)→

Original post →

More from Infra

• Nvidia's developer base hits 11 million, up ~3.7x from 3 million in 2021 — Beth_Kindig · 2026-10-16
• Muse Spark Open runs locally on a Mac M3 Ultra via llama.cpp: frontier model on desktop — pcuenq · 2026-10-16
• Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
• Meta launches Muse Spark Open, says it runs on two M5 Ultra Mac Studios with Apple's help — The Verge AI · 2026-10-16
• vLLM ships Day-0 LoRA RL for Muse Spark Open, lifting AIME from 0.56 to 0.85 — vllm_project · 2026-10-16
• DSpark draft model for Muse Spark Open proposes 7 tokens per pass with 23-42% acceptance — vllm_project · 2026-10-16

Infra channel →