🔍 Read the full analysis: Muse Spark Open: A New Option For Open-Weight AI Users 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, placing the weights of its 816-billion-parameter model on Hugging Face. Artificial Analysis scored it 38.9 on its Intelligence Index, while its results vary sharply by task: it leads tested open-weight models on two cited measures, but scored 8% on Terminal-Bench 4.0. The weights are available, but hosting requirements, licensing conditions and paid-provider costs affect how users can adopt it.
The model has 42 billion active parameters per token and 256 experts, with eight active for each token. Meta lists a one-million-token context window and text, image, speech/audio and video input; its listed output is text. The weights are offered in BF16 and NVFP4 formats on Hugging Face.
Meta’s model card estimates self-hosting needs of 700 GB of GPU memory for NVFP4 and 2 TB for BF16. It also says the model is optimized to run on two Mac Studios with M5 Ultra using MLX. Those requirements mean that access to downloadable weights does not by itself make local deployment practical for most users.
Artificial Analysis tested the model at its default xhigh reasoning effort. It reports 38.9 on Intelligence Index v4.3.2, rounded to 39 in the source report. The same report says four third-party providers listed a median price of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. It says those list prices match Meta’s proprietary Muse Spark 1.3.
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
A Large Model With Trade-Offs
Muse Spark Open gives developers access to a large model’s weights rather than only a hosted, closed service. That may matter for teams seeking control over deployment or wanting to examine and adapt an open-weight model. But the hardware estimates are substantial, and the source describes the model’s license as conditional, so the release should not be read as unrestricted, low-cost access.
Its benchmark results point to specific strengths rather than uniform leadership. Artificial Analysis reports 24.7 on AA-Omniscience, the highest score among 167 open-weight entries in that test, and a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models it compared. On visual reasoning, it scored 81% on MMMU-Pro, first among 66 open-weight entries tested.
Those results could make the model worth evaluating for visual and long-document research tasks, especially given its million-token context window. However, benchmark results do not establish performance for every workload. In particular, the reported 8% on Terminal-Bench 4.0 is a caution for users considering autonomous coding or shell-based agents.
high performance GPU for AI training
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How It Ranks Among Open Models
Muse Spark Open is the first open-weight release in Meta’s Muse Spark line. The source compares its score with Meta’s Muse Glimmer, released in August 2026, which scored 17 on the same current index, and Llama 4 Maverick, which scored 9 in the cited comparison. These results make the new release a substantial score increase over those earlier Meta models, but do not put it at the top of the overall open-weight field.
According to the source’s Artificial Analysis comparison, Muse Spark Open ranks tenth among open-weight models overall; the nine models ahead of it are from Chinese labs. Xiaomi’s MiMo-V2.6-Pro leads the cited list at 46.3. Meta’s own closed Muse Spark 1.3 scored 45.1 at the same xhigh effort and list price, while Muse Wonder scored 52.4 in the source’s account.
Price and licensing complicate comparisons. The source reports an estimated $0.84 per Intelligence Index task for Muse Spark Open, compared with $0.13 for MiMo-V2.6-Pro and $0.27 for DeepSeek V4.1 Flash. Those figures come from the source’s benchmark-cost comparison, not a guaranteed charge for every user or task. The report also says the model’s license has conditions, unlike the MIT licenses it identifies for some alternatives.
““built for coding and agentic work””
— Meta, in the Muse Spark Open model card
large memory GPU for machine learning
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License, Costs And Use Cases
The source says Muse Spark Open has a license with conditions, but the material provided does not spell out those conditions. Readers should check the Hugging Face model card and license text before commercial use, redistribution or other deployment. The exact terms and permitted uses cannot be established from the source excerpt alone.
The reported provider prices are median list prices from four third-party services, not a universal price or a promise that the same rates will remain available. Actual bills depend on provider terms and usage. The source excerpt also ends while discussing token consumption, so it does not provide a complete account of how input-token use compares across all models.
Benchmark scores are measurements under specified test settings, not guarantees of real-world results. The material does not establish how the model performs across a broad range of production deployments, nor whether the reported results will hold as providers update implementations. The 8% Terminal-Bench 4.0 score and Meta’s stronger figures on other agent tests show why task-specific evaluation matters.
As an affiliate, we earn on qualifying purchases.
Check Terms Before Deployment
Developers considering the model should review the Hugging Face weights and model card for the full license, hardware guidance and supported formats, then test it on their own workloads. Teams comparing hosted access should verify current provider pricing rather than assuming the listed median rates will apply to their account or remain unchanged.
The next useful evidence will be further independent testing and real-world reports on deployment costs, reliability and performance across different tasks. For teams building coding agents, the reported Terminal-Bench 4.0 result warrants particular scrutiny; for visual or long-context work, the model’s reported test results provide a reason to run task-specific trials, not a substitute for them.
As an affiliate, we earn on qualifying purchases.
Key Questions
When was Muse Spark Open released?
Meta released Muse Spark Open on October 15, 2026, with weights available on Hugging Face, according to the source report.
What score did it receive on the Artificial Analysis Intelligence Index?
Artificial Analysis scored it 38.9 on Intelligence Index v4.3.2, at the xhigh reasoning setting. The source report rounds that result to 39.
Can users run Muse Spark Open on their own hardware?
Meta’s model card estimates 700 GB of GPU memory for NVFP4 or 2 TB for BF16. It also says the model is optimized for two Mac Studios with M5 Ultra using MLX. These are substantial requirements, and the available source does not independently verify the estimates.
Is the model free to use?
The weights are available for download, but the source says the license has conditions and does not provide their full wording. Hosted access through third-party providers is paid; the source reports median list prices of $1.25 per million input tokens and $4.25 per million output tokens across four providers.
Is Muse Spark Open a strong choice for coding agents?
The evidence in the source is mixed. Meta describes it as built for coding and agentic work, but Artificial Analysis measured 8% on Terminal-Bench 4.0, while other reported agent benchmarks produced higher results. Teams should test their own workflows before relying on it for autonomous coding.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
