Muse Spark Open
- ID
- 35992
- Status
- summarized
- Published
- 16 Oct 2026, 1:47 AM
- Fetched
- 16 Oct 2026, 3:46 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://research.meta.ai/blog/muse-spark-open
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.0
- Created
- 16 Oct 2026, 3:46 AM
- Tags
- Audience
- developersvibe_codersai_ml_learnersai_agent_userssaas_startup_founders
What happened
Meta AI Research released Muse Spark Open, an open-weight natively multimodal model with 816B total / 42B active parameters and a 1M-token context window, targeted at coding and long-horizon agentic work. Weights are downloadable via Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden, and Microsoft Foundry Models, with APIs on Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter, and Together AI; it can also run locally on two Apple M5 Ultra Mac Studios via MLX. The Hacker News thread drew 270 points and 171 comments.
Why it matters
If you use coding agents or self-host open models, this adds a 1M-context, 42B-active option across many inference providers and stacks, but local desktop use requires two M5 Ultra Mac Studios with up to 512 GB each—not a laptop setup. For Malaysian teams, the text gives no SEA region, pricing, or data-residency detail, so the practical move is to test it through an API provider or compare it against your current model before budgeting for local hardware.
Discussion angle
Compare the real cost of self-hosting this on two M5 Ultra Mac Studios versus calling it through OpenRouter, Together, or Groq for an agentic coding workflow—where does the 1M context and 42B-active size actually pay off?