🔍 Read the full analysis: What To Know About Muse Spark Open And Its Meta-Like Pricing on ThorstenMeyerAI.com
Get the little things that make your day 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, making the weights of its first open-weight model in the Muse Spark line available on Hugging Face. Artificial Analysis gave it a score of 38.9, but the model’s conditional license, substantial self-hosting requirements and token prices matching Meta’s closed Muse Spark 1.3 complicate the meaning of “open.”
Meta released Muse Spark Open on October 15, 2026, publishing the model weights on Hugging Face and opening its flagship Muse Spark line to external use in an open-weight form. Artificial Analysis measured a score of 38.9 on its Intelligence Index, but the model is not a simple free alternative to Meta’s hosted offering: its license has conditions, self-hosting requires substantial memory, and third-party API prices match the company’s closed Muse Spark 1.3.
Muse Spark Open is a mixture-of-experts model with 816 billion total parameters and 42 billion active per token. Its model card lists 256 experts, with eight active for each token, and a 1-million-token context window. It accepts text, images, speech or audio, and video, and produces text. BF16 and NVFP4 versions of the weights are available on Hugging Face.
Meta’s model card estimates that self-hosting requires 700 GB of GPU memory for NVFP4 or 2 TB for BF16. It says the model is optimized to run on two Mac Studios equipped with M5 Ultra chips using MLX. Those requirements make the published weights accessible for developers to download, but do not mean most users can run the model locally on ordinary hardware.
Artificial Analysis tested the model at its default “xhigh” reasoning effort and reported 38.9 on Intelligence Index v4.3.2, rounded to 39 in its presentation. The source report says it ranked tenth among open-weight models overall and was the highest-scoring open-weight model from a US lab. Meta’s closed Muse Spark 1.3 scored 45.1 at the same reasoning effort, according to the same report.
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
The Cost of Using Muse Spark Open
The release matters because it pairs a high-scoring US-developed model with downloadable weights, while leaving a practical question for teams: whether they can use it under the license and afford to operate it. Open weights do not mean unrestricted use, and they do not remove the hardware or API bill.
The report lists a median API price from 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. Those are the same listed rates as Meta’s proprietary Muse Spark 1.3. The weights can be downloaded, but customers using the listed hosted services pay the same per-token rates as for the closed model.
On Artificial Analysis’s weighted cost per Intelligence Index task, Muse Spark Open cost $0.84. The report compares that with $0.13 for MIT-licensed MiMo-V2.6-Pro, which scored 46.3, and $0.27 for MIT-licensed DeepSeek V4.1 Flash, which scored 39.5. These figures are tied to the report’s specific benchmark and provider pricing; they are not universal estimates for every workload. Still, they show why organizations should compare task costs and licensing terms rather than judge the model only by its score.
There are positive results in particular areas. Artificial Analysis reported a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models it assessed, and an 81% result on its visual reasoning test, MMMU-Pro. Those metrics may make the model worth testing for research and document tasks, but they do not establish that it will be reliable for every application.
high memory GPU for AI model hosting
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A New Scale for Meta’s Open Models
Muse Spark Open is the first open-weight release from Meta’s Muse Spark model line. The report contrasts it with Muse Glimmer, a 30-billion-parameter model released in August 2026 under Apache 2.0, which scored 17 on the same current index. It also cites Llama 4 Maverick, released in April 2025, at 9. These comparisons describe the specific versions and index results covered in the source report.
The release changes the scale of Meta’s open offerings, but it does not put the model at the top of the overall open-weight rankings. Artificial Analysis listed MiMo-V2.6-Pro at 46.3, ahead of Muse Spark Open, and reported that nine open-weight models scored higher. The source says those nine were from Chinese labs. The figures therefore point to a distinction: Meta’s model leads the cited US open-weight group, while several models from other labs score higher overall.
Performance also varies by task. The report says Muse Spark Open reached 89.2 on MCP-Atlas and 80.4 on OSWorld Verified, while Artificial Analysis recorded 76% on Terminal-Bench 2.1. On the harder Terminal-Bench 4.0 test, however, it measured 8%. The report gives 33% for Muse Spark 1.3, 35% for MiMo-V2.6-Pro and 42% for GLM-5.3 on that test. The mixed results are a reason to assess coding and agent tasks separately from general benchmark performance.
““built for coding and agentic work””
— Meta’s model card
License Terms and Real-World Costs
The source material does not provide the full text of Muse Spark Open’s license, so the exact restrictions on commercial use, redistribution, modification or deployment cannot be established from the information here. The model should not be described as permissively licensed without checking the license itself. Downloadable weights and an open-weight label do not by themselves settle what uses are allowed.
API prices are reported as a median across four third-party providers, rather than a single universal price. Actual costs will depend on provider, workload, caching and token volumes. The reported $0.84 per benchmark task is likewise a result for Artificial Analysis’s weighted evaluation, not a forecast for an organization’s own use.
The benchmark figures are measurements, not guarantees of real-world accuracy or performance. The source report gives different results across agent tests and compares the model with selected alternatives; it does not establish how the model performs across every coding environment, document workflow or deployment. The release’s long-term availability, further provider support and changes to pricing or license terms are also not specified.
Check Terms Before Building
Developers considering the model can access its published weights on Hugging Face and review Meta’s model card and license before downloading or deploying it. Teams should confirm that their intended use is permitted, then test the model on representative tasks rather than relying on a single overall benchmark score.
For hosted use, buyers can compare providers’ current rates and measure input, cached-input and output-token consumption against their own workloads. For local deployment, they will need to assess whether their available hardware can meet the model’s stated memory needs. More evidence about practical costs and performance will depend on provider offerings, independent testing and results from deployments beyond the benchmarks cited in the release report.
Key Questions
What is Muse Spark Open?
It is Meta’s first open-weight model in the Muse Spark line, released on October 15, 2026. It has 816 billion total parameters, 42 billion active per token, and a 1-million-token context window.
Is Muse Spark Open free to use?
The weights are available to download, but the source report says the license has conditions and does not provide its full terms. Hosted API use is priced by the token, with reported median rates matching those of Meta’s closed Muse Spark 1.3. Check the license and provider prices for the intended use.
How did it score against other open-weight models?
Artificial Analysis reported a score of 38.9 on its Intelligence Index v4.3.2. The source report says that was the highest score among open-weight models from US labs, but nine open-weight models overall scored higher, led by MiMo-V2.6-Pro at 46.3.
Can it run on a standard computer?
Meta’s model card estimates self-hosting needs of 700 GB of GPU memory for NVFP4 or 2 TB for BF16. It also describes an optimized setup using two Mac Studios with M5 Ultra chips and MLX. These requirements are far beyond a typical personal computer.
Is it suited to autonomous coding agents?
The results cited in the report are mixed. Muse Spark Open scored 76% on Terminal-Bench 2.1, but 8% on the harder Terminal-Bench 4.0 test. Those figures do not establish how it will perform in a specific coding environment; teams should test their own tasks before relying on it.
Source: ThorstenMeyerAI.com
NFL season / tailgating Picks
team gear
As an affiliate, we earn on qualifying purchases.
