Muse Spark Open
Muse Spark Open is Meta's open-weights release of Muse Spark: an 816B-parameter (42B active) multimodal model with a 1M-token context, built for coding and agentic work.
Why it matters
Meta's open-weights release of Muse Spark, a very large multimodal model for coding and agentic work with a 1M-token context. It's for teams that want frontier-class open weights and have the hardware to run them, or a provider that does.
Highlights
- 816B total / 42B active parameter mixture-of-experts with a Perception Encoder; 1M-token context at inference.
- Input: text, image, video and audio; output: text. Trained on data from more than 100 languages.
- Meta-reported scores include Terminal-Bench 2.1 76.4, GPQA Diamond 89.0 and OSWorld Verified 80.4 at xhigh reasoning effort.
- Self-hosting needs at least 700 GB VRAM (NVFP4) or 2 TB (BF16). Supported frameworks are SGLang, vLLM, Transformers, TensorRT-LLM, Unsloth and Llama.cpp, plus MLX on two M5 Ultra Mac Studios.
- Has extra safety post-training for standalone use; Meta recommends deploying it with additional system guardrails.
Taken from the project's own README, docs or launch page. Claims about speed or accuracy are the maker's unless stated otherwise.
Worth knowing: Not intended for users under 18 per the license. Benchmarks are Meta's own.
Quick facts
| Pricing | Weights free (gated access on Hugging Face); API through third-party inference providers (prices not checked)not verified |
|---|---|
| Free or paid | Open weights (free) |
| License | Muse Model License: permissive, but requires "Built with Muse Spark Open" attribution, and Model-as-a-Service providers over $20M revenue need a separate agreement |
| Language | Model |
| Platform | Self-host or third-party API (Baseten listed on Hugging Face) |
| GitHub stars | n/a (not a GitHub project) |
| Maker | Meta Superintelligence Labs |
Links
- Official site: https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B
- Repository: https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B
Related ranking
See how established tools compare in Best Chat Assistants. New releases aren't scored until we review them against the rubric.
What we couldn't verify
- quick_facts.pricing (third-party API)
Checked 2026-10-17 from the model's own pages. Prices and star counts change, so confirm on the official site. Listing here is not a ranking or an endorsement.
The week's best new AI tools, ranked
One email a week: new releases worth your time, ranking changes, and one deep-dive. No spam.