🔍 Read the full analysis: Muse Spark Open Brings Open-Weight AI Into Meta’s Pricing Territory 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, with model weights on Hugging Face. Artificial Analysis scored it 38.9, while API providers listed the same per-token prices as Meta’s higher-scoring closed Muse Spark 1.3. The model performed strongly on hallucination and visual reasoning tests, but scored poorly on a harder terminal-agent benchmark; its license has conditions.
Meta released Muse Spark Open on October 15, publishing the weights for an 816-billion-parameter model on Hugging Face. Artificial Analysis scored it 38.9 on its Intelligence Index, while API providers listed token prices matching Meta’s closed Muse Spark 1.3, which scored higher on the same index.
Muse Spark Open is a mixture-of-experts model with 42 billion active parameters per token and 256 experts, eight of which are active for each token. It accepts text, images, speech and video as input and produces text. Meta lists the context window at 1 million tokens, and provides weights in BF16 and NVFP4 formats.
Artificial Analysis tested the model at its default xhigh reasoning effort and reports an Intelligence Index score of 38.9, rounded to 39. The outlet ranks it tenth among open-weight models overall and says it is the highest-scoring open-weight model from a US lab. Nine models ahead of it in the overall ranking are from Chinese labs.
Meta’s model card estimates self-hosting requires about 700 GB of GPU memory for NVFP4 weights or 2 TB for BF16, and says the model is optimized to run on two Mac Studios with M5 Ultra using MLX. Four third-party API providers listed median prices of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. The source says those rates match Muse Spark 1.3’s list prices.
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
Strong Reasoning, Costly API Use
The release gives developers a large Meta model whose weights can be downloaded, but using it through an API does not bring a lower token price than Meta’s closed Muse Spark 1.3, according to the listed provider rates. The source reports Muse Spark 1.3 scores 45.1 on the same index, compared with 38.9 for the open model. The weights are available without a purchase price, while self-hosting still requires substantial memory capacity.
Performance is uneven across tasks. Artificial Analysis reports a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models it tested, and an 81% result on MMMU-Pro visual reasoning, the top result among 66 open-weight entries. Those results, together with the long context window, may suit document and research workloads. They do not establish performance for every use case.
For terminal-based agents, the reported result is much weaker: Muse Spark Open scored 8% on Terminal-Bench 4.0. Artificial Analysis reports 33% for Muse Spark 1.3, 35% for MiMo-V2.6-Pro and 42% for GLM-5.3. The source also calculates a weighted cost of $0.84 per Intelligence Index task, with $0.47 attributed to Terminal-Bench 4.0. Buyers weighing the model for agent workloads therefore face both a performance question and a cost question.
As an affiliate, we earn on qualifying purchases.
Meta’s Open Model Lineup
Muse Spark Open is described in the source as the first open-weight release in Meta’s flagship Muse Spark line. Meta’s recent open releases were smaller: Muse Glimmer, a 30-billion-parameter model released in August 2026 under Apache 2.0, scored 17 on the current Intelligence Index, while Llama 4 Maverick from April 2025 scored 9, according to the source.
The release changes the scale of Meta’s open-weight offerings, but the source’s benchmark comparison does not put it at the top of the open field. Xiaomi’s MIT-licensed MiMo-V2.6-Pro scored 46.3. Other models reported above Muse Spark Open include GLM-5.3, Step 5 Preview, Kimi K3, GLM-5.3-Flash, Ling 3.1 Flash, two Qwen3.8 models and DeepSeek V4.1 Flash.
The license also affects comparisons. The source describes Muse Spark Open as having conditions in its license, while MiMo-V2.6-Pro and several other compared models use MIT licenses. Meta’s model card and the full license terms should be consulted by anyone assessing deployment rights; the source does not detail the conditions.
As an affiliate, we earn on qualifying purchases.
License Terms and Workloads
The source says the license has conditions but does not specify them. The precise obligations and restrictions, including how they apply to particular commercial deployments, cannot be determined from the supplied material alone.
Benchmark scores describe the tests and settings reported by Artificial Analysis; they do not settle how the model will perform on a developer’s own data or tasks. The source does not provide a full account of provider availability, service limits or how prices may change. Its discussion of input-token costs is also incomplete, so no broader comparison of total input use across models can be made from the supplied figures.
Open-weight AI models for developers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Check Terms Before Deployment
Developers can access the weights on Hugging Face and review Meta’s model card for deployment requirements and license terms. Those evaluating hosted use can compare provider pricing and availability directly, since the reported rates come from four third-party providers.
Further assessment will depend on testing the model on the intended workload, especially for terminal agents, and checking whether the license permits the planned use. The supplied source gives no announced follow-up release date or next benchmark milestone.
AI model training and deployment hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Muse Spark Open?
It is Meta’s 816-billion-parameter mixture-of-experts model, with 42 billion active parameters per token. Its weights are available on Hugging Face.
How did Muse Spark Open score?
Artificial Analysis gave it a score of 38.9, rounded to 39, on its Intelligence Index v4.3.2 after testing at default xhigh reasoning effort.
Is its API cheaper than Muse Spark 1.3?
The source reports that the median list prices from four third-party providers match Muse Spark 1.3’s rates: $1.25 per million input tokens, $0.15 for cached input and $4.25 per million output tokens. It says Muse Spark 1.3 scored 45.1, compared with 38.9 for the open model.
Is Muse Spark Open suitable for terminal agents?
Artificial Analysis reported an 8% score on Terminal-Bench 4.0. That result is much lower than several comparison models in the source; developers should test the model on their own agent tasks before choosing it.
What does it take to run the model locally?
Meta’s model card estimates about 700 GB of GPU memory for NVFP4 weights or 2 TB for BF16. The source also says Meta describes an optimized setup using two Mac Studios with M5 Ultra and MLX.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
