🔍 Read the full analysis: A Guide To Muse Spark Open’s Open-Weight Approach And Price on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Meta released Muse Spark Open on October 15, 2026, making the first open-weight model in its flagship Muse Spark line available on Hugging Face. Artificial Analysis scored it 38.9 and found strengths in hallucination control and visual reasoning, but its hosted API list price matches the higher-scoring closed Muse Spark 1.3, and the license has conditions.
Meta released Muse Spark Open on Hugging Face on October 15, 2026, its first open-weight model in the flagship Muse Spark line. Artificial Analysis scored the 816-billion-parameter model at 38.9 on its Intelligence Index, while provider list prices match those for the closed Muse Spark 1.3, which scored higher at 45.1.
Muse Spark Open is a mixture-of-experts model with 816 billion total parameters and 42 billion active per token. Meta offers weights in BF16 and NVFP4 formats. Its model card estimates that self-hosting requires 700 GB of GPU memory for NVFP4 or 2 TB for BF16, and says the model is optimized to run across two Mac Studios with M5 Ultra using MLX.
The model accepts text, images, audio and video, produces text, and has a one-million-token context window. Artificial Analysis tested it at its default xhigh reasoning effort. The index score of 38.9 is rounded to 39 in the source report. Artificial Analysis places it tenth among open-weight models overall and says it is the highest-scoring open-weight model from a US lab.
The weights can be downloaded, but hosted access carries a bill amid open-weight AI price battles. Across four third-party providers, the reported median list prices are $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. Those figures match the reported list prices for closed Muse Spark 1.3. A weighted Intelligence Index task costs $0.84 on Muse Spark Open, according to the source 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
A Large Model With a Split Case
The release changes the scale of Meta’s open-weight offering: at 816 billion parameters, Muse Spark Open is much larger than recent company releases cited in the source, including Muse Glimmer at 30 billion. Its score also leads the US open-weight models listed in the report. But it does not lead the overall open-weight field, where the cited leaders are from Chinese labs.
Its strongest results point to particular uses. Artificial Analysis measured a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models in its comparison, and an 81% score on MMMU-Pro visual reasoning, the top result among 66 open-weight entries tested. Those results, alongside a long context window, may appeal to teams handling documents and research. The report also cautions that the model performed poorly on Terminal-Bench 4.0, a benchmark of terminal-based agent work, with an 8% score.
For developers, access and price are separate decisions. Downloading weights does not require paying a per-token API rate, but self-hosting demands substantial memory. Those using a provider face the same listed token prices as for Muse Spark 1.3, despite its higher index score. For one benchmark task, the reported cost was also well above several open-weight alternatives cited in the source.
high performance GPU for AI hosting
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Meta’s Open Model Lineup
Muse Spark Open is the first open-weight release in Meta’s Muse Spark model family. The source report compares it with Muse Glimmer, released in August 2026, which scored 17 on the same index, and Llama 4 Maverick from April 2025, which scored 9. It also compares the model with NVIDIA’s Nemotron 3.5 Super VL, released on the same day and scored at 26.5.
The overall rankings give a different view from the US-only comparison. The report lists Xiaomi’s MIT-licensed MiMo-V2.6-Pro at 46.3, above Muse Spark Open’s 38.9, and identifies nine open-weight models ahead of Meta’s release. It also compares Muse Spark Open with Meta’s closed Muse Spark 1.3, which scored 45.1 at the same reasoning effort and list price. The source says Muse Wonder scored 52.4, though it does not provide a direct comparison of its price here.
Artificial Analysis reported mixed task results. Muse Spark Open scored 89.2 on MCP-Atlas, 80.4 on OSWorld Verified and 76% on Terminal-Bench 2.1. On the harder Terminal-Bench 4.0, however, it scored 8%; the cited comparison scores were 33% for Muse Spark 1.3, 35% for MiMo-V2.6-Pro and 42% for GLM-5.3.
““built for coding and agentic work””
— Meta, in the Muse Spark Open model card
License Terms and Real-World Costs
The source report describes the license as having conditions, but the supplied material does not spell out those terms. Readers considering deployment need to review the license itself to determine what uses it permits and what obligations apply. The available information also does not establish how costs or performance will vary across providers, prompts and production workloads.
The benchmark results are measurements under specified tests, not guarantees of performance in other tasks. The report gives a weighted cost for an Intelligence Index task and token prices, but the supplied material does not provide a full account of every workload assumption behind those figures. The practical cost of a given application will depend on its input, cached and output token use.
Deployment Choices Ahead
Developers can download the BF16 or NVFP4 weights from Hugging Face, assess whether their hardware meets Meta’s stated memory requirements, or use one of the third-party providers listed for API access. Teams considering either route will need to review the model license and test performance against their own tasks, particularly if they plan to use terminal-based agents.
The source material does not name a further release date or a scheduled follow-up from Meta. The next practical evidence will come from provider pricing and independent tests as users evaluate the model in their own workloads; any changes to the model, license or availability remain unconfirmed in the material provided.
Key Questions
When did Meta release Muse Spark Open?
Meta released it on October 15, 2026. The weights are available on Hugging Face, according to the source report.
Is Muse Spark Open free to use?
The weights are available to download, but the source says the license has conditions and does not specify them. Hosted API use has per-token prices, with a reported median of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens.
How did Muse Spark Open score?
Artificial Analysis measured 38.9 on its Intelligence Index v4.3.2 at xhigh reasoning effort. The source says the index rounds that to 39.
How much hardware does self-hosting require?
Meta’s model card estimates 700 GB of GPU memory for the NVFP4 version and 2 TB for BF16. It also says the model is optimized to run on two Mac Studios with M5 Ultra using MLX.
Is Muse Spark Open suited to coding agents?
Results vary by benchmark. The source reports 76% on Terminal-Bench 2.1 but 8% on Terminal-Bench 4.0, and says the latter result is weak compared with several listed alternatives.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
