Meta releases Muse Spark Open: 816B MoE open-weight model with 1M context built for coding agents
ryanshrout · x · 2026-10-16
Meta released Muse Spark Open, billed as the most capable US open-weight model to date, targeting coding and agentic work: - 816B total / 42B active params, MoE with top-8 routing across 256 experts - 1M token context window for whole-codebase tasks - Natively multimodal (text/image/video/audio) via a 1.8B perception encoder - Trained on NVIDIA GB300 NVL72; NVFP4 checkpoints for Blackwell Ultra deliver 50K+ tokens/sec/GPU. Weights are on Hugging Face, with a single-command NVIDIA NIM deployment option.
More from coding & agent
- David Hoang: Git Worktree is the cause of, and solution to, all problems — davidhoang · 2026-10-16
- Why decomposing tasks into micro-decisions matters for AI automation — hardimanjames · 2026-10-16
- "Bribe coding": getting Claude to work by bribing it, and running three agents in parallel — jia_seed · 2026-10-16
- Vercel Connect adds MCP Events in public beta for event-driven agents — cramforce · 2026-10-16
- PE operating partner shares real agent deployments: 59/60 accuracy at $250/month — alex_verem · 2026-10-16
- Dev: live voice + tool-calling + image gen now combine into a workable 'Her'-like experience — andreisavu · 2026-10-16