Meta
Muse Spark Open
Meta's 816B-parameter open-weight mixture-of-experts reasoning model (October 2026) aimed at coding agents and tool use.
Our editorial assessment, based on the public sources listed below. Last reviewed .
- Overall rank
- #36
- Tier
- A
- Score
- 82.3
- Main category
- Coding
Strengths
- Weights published on Hugging Face and listed on SageMaker, Vertex Model Garden and Microsoft Foundry
- 1M-token context with text, image, audio and video input
- Scores above the open-weight median for its size class on Artificial Analysis, with fast output speed
Considerations
- Custom Muse Model License: attribution notice and a separate agreement for large hosted services
- Self-hosting needs roughly 700GB+ of accelerator memory; hosted API prices sit above the open-weight median
Price
Open weights (Muse Model License); hosted APIs around $1.25 input / $4.25 output per 1M tokens
Shown for reference — price is not part of the score.
Score breakdown
- Performance
- 82.0
- 55% of score
- Reputation
- 78.0
- 35% of score
- Recency
- 99.0
- 10% of score
The overall score combines these sub-scores using the weights shown. How we score →
Verification status
Each field carries its own re-verification date. Amber means older than 90 days — treat those figures as stale until re-checked.
Checked against the sources listed below.
Sources
- Artificial Analysis — Muse Spark Open (opens in a new tab)Accessed:
- ITdaily — Meta makes Muse Spark available as an open-weight model (opens in a new tab)Accessed:
- Hugging Face — Muse Spark Open model card (Meta) (opens in a new tab)Accessed:
- Artificial Analysis — LLM Leaderboard (opens in a new tab)Accessed:
Similar models
- GPT-5.3 CodexOpenAI#32 · Tier A · Score 82.9Compare
- Claude Mythos PreviewAnthropic#33 · Tier A · Score 82.4Compare
- Claude Opus 4.7Anthropic#37 · Tier A · Score 82.0Compare
- Claude Sonnet 4.6Anthropic#42 · Tier A · Score 81.3Compare