Muse Spark Open
Language modelOpen weightsCard summarizedBuilt for coding and agentic work, Muse Spark Open is natively multimodal.
Use it via API
Where to call it
- Self-host (weights)
meta-models/Muse-Spark-Open-816B-A42B - Together AI
meta-models/muse-spark-open - Google Vertex AI
—
Good for
- Multi-step reasoning across long, complex workflows
- End-to-end agentic tasks: planning, sequential tool calls, failure recovery
- Coding agents: writing, debugging and resolving real software tasks
Overview
Spark Open
Built for coding and agentic work, Muse Spark Open is natively multimodal.
- Cost
- $1.25 host
- Context
- 1M
- Cheapest host
- $1.25 / $4.25
- Max output
- —
- Knows up to
- —
- Released
- 15 Oct 2026
The open-weights release of Muse Spark: an 816B-total, 42B-active mixture-of-experts model with a 1M-token context, built for coding and agentic work and natively multimodal. Meta added extra safety post-training so it holds up as a standalone model without system guardrails, and says it met the release thresholds of its Superintelligence Scaling Framework. Meta does not say which Muse Spark version it matches. model card; research blog
meta-models/Muse-Spark-Open-816B-A42B- Input / output per 1M
- $1.25 / $4.25 via Together AI
- Context window
- 1M
- Max output
- —
- Knows up to
- —
- Reasoning
- Reasoning effort low / medium / high / xhigh
- Weights
- Open · Muse Model License
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Meta's Muse Spark Open model card (Hugging Face). Every claim carries its source page.