Meta open-sources Muse Spark Open: 816B MoE with 42B active, 1M context, strong benchmarks
testingcatalog · x · 2026-10-16
Meta released Muse Spark Open (Muse-Spark-Open-816B-A42B) on Hugging Face: an open-weights mixture-of-experts model with 816B total / 42B active parameters and a 1M-token context window, built for agentic and coding workflows. BF16 and NVFP4 weights are available. Shared benchmarks show it leading peers on MCP-Atlas (89.2), HLE-Diamond (26.1) and AA-Omniscience (25), while trailing Kimi K3 on Terminal-Bench 2.1 (76.4 vs 85.0).
More from Models
- Nathan Lambert: new US open-weight release cuts gap with Chinese models by half — natolambert · 2026-10-16
- Swiss AI Initiative releases 134-page Apertus 1.5 technical report on fully open multimodal reasoning — valentina__py · 2026-10-16
- NVIDIA Nemotron 3.5 Super VL lands in Perplexity for Pro/Max and Agent API — perplexity_ai · 2026-10-16
- Meta releases Muse Spark Open 816B-A42B weights on Hugging Face — mishig25 · 2026-10-16
- Meta's 816B open-weight Muse Spark Open hits Groq preview at 350+ tokens/sec — GroqLLC · 2026-10-16
- LLMs are 7.5x more likely to tell you not to change, Waterloo study finds — alex_verem · 2026-10-16