Meta releases 816B-parameter open-weight Muse Spark Open with 1M context for agentic coding
woosuk_k · x · 2026-10-16
Meta has released Muse Spark Open, an open-weight model that vLLM is supporting from day one, calling it the most capable US open-weight model.
- Scale: 816B total parameters, 42B active per token
- Context: 1M tokens with native image, video, audio and document understanding, built for long-horizon agentic coding on large codebases
- Training: vLLM powered RL inference during training; vLLM contributor zhuohan123 frames it as recursive self-improvement — RL improves models, LLMs optimize inference, faster inference enables more RL
- Deployment: BF16, NVFP4 & MXFP4 support, DSpark speculative decoding on CUDA and ROCm, verified on NVIDIA and AMD GPUs plus TPUs
Related event: Meta Open-Sources Muse Spark Open, an 816B-Parameter Multimodal MoE Model(81 posts)→
More from coding & agent
- Terminal Night 3 talk slides released, missing demo video to follow — 4310sy · 2026-10-16
- Agent developers are sitting on unused API credits that can exceed actual usage — EnvironmentalDay8864 · 2026-10-16
- Notion's bet on agents: abstract away deployment so users and agents just write code — simonguozirui · 2026-10-16
- Four Gemini 4 Argon agents beat Minecraft's Ender Dragon autonomously in 27 hours — Maxence Faldor · 2026-10-16
- Tencent's Terminal-RSI builds recursively self-improving terminal agents, hitting 19.84% on Terminal-Bench 4.0 — tencent · 2026-10-16
- Tsinghua's AutoSFT uses multi-agent collaboration to automate SFT data research — TsinghuaNLP · 2026-10-16