🔍 Read the full analysis: How Much Does Open-Weight Strength Cost? Muse Spark Open’s Price Compared 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 weights on Hugging Face. Artificial Analysis measured a score of 38.9 and strong results in hallucination and visual reasoning, but its $0.84 weighted cost per benchmark task matches the listed token prices for closed Muse Spark 1.3, which scored higher. Its license has conditions, and its results on a harder agent benchmark were weak.
The model is a mixture of experts with 42 billion active parameters per token, drawing on 8 of 256 experts. Meta lists a one-million-token context window and support for text, images, audio and video input, with text output. The weights are available in BF16 and NVFP4 formats. Meta’s model card estimates self-hosting requires about 2 TB of GPU memory in BF16 or 700 GB in NVFP4, and says the model is optimized for two Mac Studios with M5 Ultra chips using MLX.
Artificial Analysis tested Muse Spark Open at its default xhigh reasoning effort. It reports an Intelligence Index v4.3.2 score of 38.9, rounded to 39. The index places it tenth among open-weight models overall and first among US open-weight models. The nine models ranked above it are from Chinese labs. Its score is below closed Muse Spark 1.3’s 45.1 under the same effort setting; Muse Wonder scored 52.4, according to the source material.
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. The source reports these prices match Muse Spark 1.3’s list prices. In Artificial Analysis’s benchmark accounting, Muse Spark Open cost $0.84 per Intelligence Index task. That is above MIT-licensed MiMo-V2.6-Pro’s $0.13 and DeepSeek V4.1 Flash’s $0.27; the latter scored 39.5, compared with Muse Spark Open’s 38.9.
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
Performance Comes With a Cost
Muse Spark Open expands the range of high-capability US open-weight models, but its benchmark position alone does not settle whether it is a practical choice. Buyers must weigh provider charges, hardware needs and license terms alongside its scores. The weights can be downloaded, but serving the model still requires substantial infrastructure or paid access through a provider.
Its performance is uneven across tasks. 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 also scored 81% on the visual reasoning benchmark MMMU-Pro, leading the 66 open-weight models tested there. These results may suit research and document workflows, though benchmark scores do not establish performance for every real-world use.
For agentic work, the results vary by test. The model scored 76% on Terminal-Bench 2.1, but 8% on the harder Terminal-Bench 4.0 benchmark used in the Intelligence Index. Artificial Analysis reports the 4.0 result accounted for $0.47 of the $0.84 weighted task cost. That combination makes the model’s performance and expense particularly relevant to teams evaluating long-running automated tasks.
Popular right now
Hisense 40-Inch Class QD4SR Series Hi-QLED FHD Smart Roku TV (2026 Model)
De'Longhi La Specialista Touch Espresso Machine with Coffee Grinder
Samsung 75-Inch The Frame Pro 4K Neo QLED Smart TV (QN75LS03FW)
As an affiliate, we earn on qualifying purchases.
From Meta’s Smaller Open Models
Muse Spark Open is the first open-weight release in Meta’s Muse Spark family. The source compares it with Muse Glimmer, a 30-billion-parameter model released in August 2026 under Apache 2.0, which scored 17 on the current index, and Llama 4 Maverick from April 2025, which scored 9. The new model substantially raises the scale and benchmark standing of Meta’s open offerings.
It does not lead the full open-weight field on the Intelligence Index. Artificial Analysis scored MiMo-V2.6-Pro at 46.3, with GLM-5.3, Step 5 Preview, Kimi K3 and several other models also ahead of Muse Spark Open. NVIDIA’s Nemotron 3.5 Super VL, released the same day, scored 26.5, according to the source. The comparison is not solely about rankings: some competitors have permissive MIT licenses, while Muse Spark Open’s license includes conditions.
License and Use Limits
The supplied material describes the Muse Spark Open license as having conditions but does not specify those terms. Readers should review Meta’s license and model card before deploying or redistributing the model. It is also not clear from the supplied information how the reported provider median prices may change, or whether provider implementations will match the benchmark setup.
Benchmark results describe defined tests, not guaranteed outcomes in production. The material gives selected agent and reasoning scores, but does not establish how the model performs across a particular organization’s workloads, infrastructure or evaluation methods. Self-hosting estimates are attributed to Meta’s model card; actual memory requirements may depend on deployment configuration.
Check Terms Before Deployment
Teams considering Muse Spark Open can download its weights from Hugging Face and compare the license, hardware requirements and provider rates with their intended workload. The next useful step is to evaluate it on representative tasks, with particular attention to input-token costs and agent runs, which contributed substantially to the reported benchmark bill.
Further provider listings, independent evaluations and details on the license’s practical conditions could clarify how the model fits into production. Until then, the available results show a high-scoring US open-weight release with standout scores on some tests, alongside substantial serving costs and weaker results on a harder terminal-agent benchmark.
Key Questions
When did Meta release Muse Spark Open?
Meta released Muse Spark Open on October 15, 2026, with weights available on Hugging Face.
How did Muse Spark Open score?
Artificial Analysis measured a score of 38.9 on its Intelligence Index v4.3.2 at default xhigh reasoning effort. The source says this placed it tenth among open-weight models overall and first among US open-weight models.
How does its price compare with similar models?
Artificial Analysis reported a weighted cost of $0.84 per benchmark task. In the supplied comparison, MIT-licensed MiMo-V2.6-Pro cost $0.13 per task and scored 46.3; DeepSeek V4.1 Flash cost $0.27 and scored 39.5.
Is Muse Spark Open permissively licensed?
The supplied source says its license has conditions, but does not detail them. Check Meta’s license terms before using, modifying or redistributing the weights.
Is it strong at coding and agent tasks?
Results depend on the benchmark. It scored 76% on Terminal-Bench 2.1, while Artificial Analysis measured 8% on Terminal-Bench 4.0, the harder version used in its Intelligence Index.
Source: ThorstenMeyerAI.com
