Meta's Muse Spark Open tops US open-weights models on Artificial Analysis
New models on Complete
Meta released Muse Spark Open on October 15, 2026. It is an open-weights model built for coding and multi-step work, and it shares its base model with Meta's Muse Spark 1.3. Muse Spark Open is available for enterprise teams on Complete today.
The top US model you can run yourself
Artificial Analysis scores Muse Spark Open 39 on its Intelligence Index, the highest of any US open-weights model.
Open weights means you can download the model and run it on infrastructure you control, which keeps sensitive data inside your own environment. You can also use it through hosted providers.
Fewer made-up answers
Artificial Analysis tests how often models state wrong facts with confidence instead of declining to answer. Muse Spark Open has a 24% hallucination rate, the lowest of the comparable open-weights models Artificial Analysis tracks.
For document question answering, research summaries and customer-facing answers, a lower rate means fewer errors to catch.
Reads text, images, speech and video
Muse Spark Open takes text, images, speech and video as input and replies in text. It reads up to 1 million tokens in a single request.
It is also quick. Artificial Analysis measures about 103 output tokens per second across providers, more than twice the speed of GPT-6 Astra, and it starts responding in about 0.65 seconds.
How it compares on cost
Model | Intelligence Index | Cost per task |
|---|---|---|
Gemini 4 Argon (high) | 53 | $1.99 |
GPT-6.1 Sol (max) | 52 | $0.72 |
MiMo-V2.6-Pro | 46 | $0.13 |
GLM-5.3 (max) | 45 | $2.01 |
Kimi K3 (max) | 44 | $2.00 |
Muse Spark Open (xhigh) | 39 | $0.84 |
DeepSeek V4.1 Flash (max) | 39 | $0.27 |
GPT-6 Luna (max) | 38 | $0.07 |
Source: Artificial Analysis Intelligence Index v4.3.2, October 2026. Cost per task counts both the token price and how much the model writes to finish each task.
The leading Chinese open-weights models, GLM-5.3 and Kimi K3, score higher at 45 and 44. Muse Spark Open finishes a task for about 40% of their cost. DeepSeek V4.1 Flash matches its score at a lower cost per task, and MiMo-V2.6-Pro scores higher for less.
Choose Muse Spark Open when a US-built model, open weights, mixed inputs and fewer hallucinations matter to the workflow.
License terms
Muse Spark Open ships under the Muse Model License. Commercial use is subject to its conditions, including "Built with Muse Spark Open" attribution in products that use it.
Where to run it
Hosted: Groq, Together AI, Fireworks, Baseten, Nebius, Modal and OpenRouter
Cloud model catalogs: Amazon SageMaker, Google Vertex AI Model Garden and Microsoft Foundry
Weights: Hugging Face
Enterprise teams on Complete can put Muse Spark Open to work on their own workflows now.
Book a demo to run Muse Spark Open on your workflows.
Frequently asked questions
Is Muse Spark Open the best open-weights model?
It is the top-scoring US open-weights model on the Artificial Analysis Intelligence Index (score 39 at xhigh reasoning). Chinese open-weights models GLM-5.3 (45) and Kimi K3 (44) score higher on the same index.
Can I use Muse Spark Open commercially?
Commercial use is permitted under the Muse Model License, subject to its conditions. Products that use it must display "Built with Muse Spark Open" attribution.
Where can I run Muse Spark Open?
Hosted: Groq, Together AI, Fireworks, Baseten, Nebius, Modal and OpenRouter. Cloud catalogs: Amazon SageMaker, Google Vertex AI Model Garden and Microsoft Foundry. Weights are on Hugging Face for self-hosting.
Does Muse Spark Open run locally?
Yes. The weights are on Hugging Face, so you can run it on GPU infrastructure you control.
