🔍 Read the full analysis: Muse Spark Open Combines Open Weights With Meta-Level Pricing on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Meta released Muse Spark Open on October 15, 2026, with downloadable weights and an 816-billion-parameter mixture-of-experts design. Artificial Analysis rated it 38.9, while listed provider prices match Meta’s closed Muse Spark 1.3; the open model scored lower and showed weaker results on a demanding agent benchmark. Its license terms and practical deployment costs matter for teams considering it.
Meta released Muse Spark Open, an 816-billion-parameter open-weight model, on Hugging Face on October 15, 2026. Artificial Analysis measured a score of 38.9 on its Intelligence Index, while third-party providers list token prices matching Meta’s closed Muse Spark 1.3, which scored higher on the same index.
The model uses a mixture-of-experts design with 256 experts and 42 billion active parameters per token. Meta lists a one-million-token context window and support for text, images, audio and video as input, with text output. Weights are available in BF16 and NVFP4 formats. Meta’s model card estimates that self-hosting requires about 700 GB of GPU memory for NVFP4 or 2 TB for BF16, and says the model is optimized to run on two Mac Studios with M5 Ultra using MLX.
Artificial Analysis tested the model at its default xhigh reasoning effort. Its 38.9 Intelligence Index score makes Muse Spark Open the tenth-highest-scoring open-weight model in the assessment. The source report says the nine models ahead of it are from Chinese labs. Among US open-weight releases, it scores above NVIDIA’s Nemotron 3.5 Super VL, which received 26.5 in the same account, though the models’ license terms differ.
The release has notable strengths on selected measures. Artificial Analysis recorded 24.7 on AA-Omniscience, the highest score among 167 open-weight models in that test, and a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight entries. It also scored 81% on MMMU-Pro visual reasoning, leading the 66 open-weight entries tested. These results sit alongside weaker performance on Terminal-Bench 4.0: 8%, compared with 33% for Muse Spark 1.3 and higher results for several open competitors.
Infographic 01 · Where Muse Spark Open lands
The best US open-weight model is tenth. The nine ahead are Chinese.
Artificial Analysis scores Meta’s first open Muse Spark model at 38.9. That is 12 points clear of any other US open-weight model, but behind nine Chinese models. Four of those nine are MIT-licensed, and three of them cost a third or less per task.
Top open-weight models by Intelligence Index
Higher is better. Effort variants of the same model are not listed twice. Cost = average API cost to run one index task.
License: Muse Spark Open ships under the Muse Model License: broad use and redistribution, but mandatory “Built with Muse Spark Open” attribution, a separate agreement for model-as-a-service businesses above $20M revenue, and a usage policy.
Source: Artificial Analysis, Muse Spark Open (xhigh) model page and model data, read 16 Oct 2026. Intelligence Index v4.3.2, ten evaluations. License categories as listed by Artificial Analysis. Countries: Xiaomi, Z AI, StepFun, Kimi, InclusionAI, Alibaba and DeepSeek are Chinese labs; Meta and NVIDIA are US.
Powered by Thorsten Meyer AI
Open Weights, Closed-Model Rates
The release gives developers access to weights from Meta’s flagship Muse Spark line, a notable change from the company’s recent smaller open models. That may make the model relevant to teams working with long documents, visual reasoning or research tasks, especially given its one-million-token context window and results on AA-Omniscience and MMMU-Pro.
Access to weights does not mean low operating costs. Four third-party providers list a median price of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. Those prices match Muse Spark 1.3’s listed rates, according to the source report. Artificial Analysis estimated a weighted cost of $0.84 per Intelligence Index task; its comparison put MiMo-V2.6-Pro at $0.13 and DeepSeek V4.1 Flash at $0.27. Costs will depend on provider, workload and usage.
The evaluation also cautions against treating one overall score as a fit for every workload. Muse Spark Open’s results on some coding and agent benchmarks contrast with its 8% on Terminal-Bench 4.0. For teams choosing a model to run tool-using agents, that result and the per-task cost are relevant alongside its stronger tests.
GPU memory optimizer for AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Muse Spark Compares
Muse Spark Open is the first open-weight release in Meta’s Muse Spark model line. The source report compares it with Muse Glimmer, a 30-billion-parameter model released under Apache 2.0 in August 2026, which scored 17 on the current index, and Llama 4 Maverick, which scored 9 in April 2025. The newer model is much larger than those releases and substantially ahead of them on the cited index.
That does not put it at the top of the full open-weight field. Artificial Analysis’ index placed Xiaomi’s MIT-licensed MiMo-V2.6-Pro at 46.3; the source also lists several other models scoring above Muse Spark Open. Meta’s own closed Muse Spark 1.3 scored 45.1 at the same reasoning effort and list price, while Muse Wonder scored 52.4. Those comparisons come from the source report’s account of the index and should be read as benchmark results, not predictions of performance on every task.
The model’s public availability also comes with license conditions, according to the source report, while some competing models are offered under MIT or Apache 2.0. The specific terms can affect whether and how an organization may use or redistribute the weights. Prospective users need to review Meta’s license and model card directly before building on the release.
“Built for coding and agentic work.”
— Meta, in its Muse Spark Open model card
high performance GPU for machine learning
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License and Workload Fit
The supplied source material does not set out the full license conditions, so the limits on commercial use, redistribution or other deployment choices cannot be detailed here. The source says the license has conditions; users should consult Meta’s published terms before adopting the weights.
Benchmark scores do not establish how the model will perform on a particular organization’s documents, tools or hardware. The source also does not provide a complete account of provider availability, price changes or the costs of self-hosting beyond Meta’s memory estimates. It remains unclear how actual costs will vary across workloads and whether the reported benchmark gaps will carry over to production use.
As an affiliate, we earn on qualifying purchases.
Check Terms and Test Workloads
The weights are available on Hugging Face, and third-party providers list API access. Teams considering the model can review Meta’s model card and license, compare provider rates, and test it on their own documents and tasks. Those steps can show whether its long-context and visual reasoning results justify the listed token prices for a given use.
Further provider pricing, independent evaluations and real-world deployment results may clarify where Muse Spark Open fits. The source material does not identify a scheduled follow-up release or a next benchmark update.
As an affiliate, we earn on qualifying purchases.
Key Questions
When did Meta release Muse Spark Open?
Meta released the model on October 15, 2026. Its weights are hosted on Hugging Face.
What did it score on Artificial Analysis’ Intelligence Index?
Artificial Analysis measured 38.9 at its default xhigh reasoning effort; the source report says the index rounds that to 39. It ranked tenth among the open-weight models in the assessment.
How much does API access cost?
The source report gives a median list price across four third-party providers of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. Prices can vary by provider and may change.
Can organizations use the weights without restrictions?
The source report describes the license as having conditions but does not provide its full terms. Organizations should read Meta’s published license before deciding whether their intended use is allowed.
Is Muse Spark Open strong at agent tasks?
Results vary by benchmark. The source reports scores including 89.2 on MCP-Atlas and 76.4% on Terminal-Bench 2.1, but Artificial Analysis measured 8% on Terminal-Bench 4.0. That spread makes workload-specific testing useful before deployment.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
