Meta's PyTorch blog details Muse Spark Open: safety-hardened 816B coding model

lmoroney · x · 2026-10-16

Meta's PyTorch team blog announces the open-weights release of Muse Spark Open with full technical details:

• Billed as the most capable US open-weights model to date, live on Hugging Face under the Muse Model License
• 816B total / 42B active MoE with a 1M-token context window, natively multimodal, sharing the same base model as Muse Spark 1.3
• Coding-optimized for agentic workflows: higher first-attempt accuracy, reliable tool calling and deep codebase understanding, trained on real developer workflows (multi-file refactors, debugging sessions, long-running tasks)
• Extra safety post-training versus the August Muse Spark release makes it robust as a standalone model without system guardrails
• PyTorch ecosystem: Helion kernel DSL for high-performance ops, TorchTitan supporting open SFT and RL out of the box, inference optimized for SGLang/vLLM on NVIDIA, AMD and Intel hardware

Related event: Meta Open-Sources Muse Spark Open, an 816B Multimodal Model Billed as America's Strongest Open-Weight Model(85 posts)→

Original post →

More from Models

• Gemini 4 model card: model eliminated months of trial-and-error on bioweapons research — Puzzleheaded-King584 · 2026-10-16
• Every AI picks 7: Anthropic engineer's rule for escaping model taste convergence — rubenhassid · 2026-10-16
• Brain teaser: strip all attention, MLP and embedding matrices from a Llama 3 — how many trainable params remain? — Sauers_ · 2026-10-16
• Lean v4.35.0 ships verified kernels — one written and proven consistent by Claude — geoffreyirving · 2026-10-16
• One flagged seat froze an entire Claude team org — founders share backup plans — RexDouglass · 2026-10-16
• Kimi K3 checkpoint impresses: strong held-out results point to solid pretraining — SonglinYang4 · 2026-10-16

Models channel →