🔍 Read the full analysis: Muse Spark Open: A Strong Open-Weight Model With Meta’s Price Benchmark on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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, publishing the model weights on Hugging Face. Artificial Analysis scored it 38.9 on its Intelligence Index, with the best AA-Omniscience result among open-weight models tested, while its listed API prices match Meta’s higher-scoring closed Muse Spark 1.3. Its license has conditions, and the model performed poorly on a harder terminal-agent benchmark.
Meta released Muse Spark Open, an 816-billion-parameter mixture-of-experts model, on October 15, publishing its weights on Hugging Face. Artificial Analysis scored it 38.9 on its Intelligence Index v4.3.2, making it a high-scoring US open-weight release, but its listed API rates match the more capable closed Muse Spark 1.3 and the model’s license includes conditions.
Muse Spark Open has 42 billion active parameters per token and 256 experts, with eight active for each token. It accepts text, images, speech and audio, and video, and produces text. Meta offers BF16 and NVFP4 weights; its model card estimates self-hosting requires about 2 TB of GPU memory for BF16 or 700 GB for NVFP4. Meta also says the model is optimized to run on two Mac Studios with M5 Ultra using MLX.
Artificial Analysis tested the model at its default xhigh reasoning effort. It reported a 38.9 Intelligence Index score, rounded to 39, and ranked the model tenth among open-weight models overall. The nine models ahead of it in that ranking came from Chinese labs, according to the supplied results. Artificial Analysis described Muse Spark Open as the leading US open-weight model; NVIDIA’s Nemotron 3.5 Super VL scored 26.5 on the same index.
Meta’s model weights are available at no charge, but four third-party providers listed API access at a median price of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens. Those rates match the listed prices for Meta’s closed Muse Spark 1.3. At the same reasoning effort, Spark 1.3 scored 45.1, about six points higher.
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 US Release With Trade-Offs
The release changes the scale of Meta’s open-weight offerings: Muse Spark Open is the first model in its flagship Muse Spark line to have open weights, and its 816-billion-parameter total is much larger than the company’s recent open releases cited in the source. The score also places it well above the listed US open-weight comparisons, including Muse Glimmer at 17 and Llama 4 Maverick at 9.
For developers, however, the benchmark score does not settle whether the model is a good fit. Artificial Analysis found strong results in visual reasoning and resistance to hallucinations, but weaker performance on some agentic tasks. The listed API prices also make the model less attractive on cost than several MIT-licensed open-weight competitors in the supplied comparison: MiMo-V2.6-Pro scored 46.3 for a reported $0.13 per index task, while Muse Spark Open scored 38.9 for $0.84. These figures come from Artificial Analysis’s benchmark and cost estimates, not a guarantee of costs for every workload.
The license is another factor for organizations considering deployment. The supplied source describes it as conditional, but does not reproduce its terms. Teams will need to review the license itself before deciding whether it permits their intended uses.
GPU high memory requirement for AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How It Compares Across Benchmarks
Artificial Analysis reports that Muse Spark Open leads the tested open-weight entries on two measures highlighted in the source. It scored 24.7 on AA-Omniscience, which rewards correct answers and penalizes confident wrong ones, ranking first among 167 open-weight models. Its reported hallucination rate was 24%, the lowest among the 15 highest-scoring open-weight models in that comparison. The source gives rates of 28% for GLM-5.3-Flash, 41% for MiMo-V2.6-Pro and 53% for Kimi K3.
On visual reasoning, Muse Spark Open scored 81% on MMMU-Pro, ranking first among 66 tested open-weight models. The supplied results also list 89% on GPQA Diamond, 44% on Humanity’s Last Exam and 58% on SciCode. Its 1-million-token context window and those results may interest people working with long documents or research material, though benchmark performance alone does not establish performance on a particular task.
Results on agentic tests were mixed. Meta’s model card calls the model built for coding and agentic work, and reports 89.2 on MCP-Atlas, 80.4 on OSWorld Verified and 76.4% on Terminal-Bench 2.1. Artificial Analysis reported 8% on the harder Terminal-Bench 4.0 benchmark included in its index. The source also reports 50% on AutomationBench-AA, 1,401 Elo on GDPval-AA and 1,279 Elo on AA-Briefcase. Those scores trail several listed comparisons, including Muse Spark 1.3 on GDPval-AA and AA-Briefcase.
“built for coding and agentic work”
— Meta, in the Muse Spark Open model card
As an affiliate, we earn on qualifying purchases.
License and Deployment Questions
The supplied material does not spell out the conditions in Muse Spark Open’s license, so its practical limits for commercial use, redistribution or modification cannot be established from these details alone. The model card’s hardware estimates and Mac Studio optimization statement are Meta’s guidance; independent deployment results are not provided here.
The benchmark and pricing figures are snapshots from the release day. The source does not provide enough information to assess how performance or provider rates may change, or how the model will behave across specific real-world workloads. The reported cost comparison is tied to Artificial Analysis’s index tasks and should not be treated as a universal price estimate.
As an affiliate, we earn on qualifying purchases.
Provider Access and Real-World Use
Developers can access the published weights on Hugging Face, while API access is available through the four providers listed in the supplied source. Prospective users can compare the model’s results with their own workloads, check current provider rates and review the license terms before deployment. Further evidence about practical use will depend on independent evaluations and how the model performs on tasks beyond the reported benchmark suite.
As an affiliate, we earn on qualifying purchases.
Key Questions
When did Meta release Muse Spark Open?
Meta released it on October 15, 2026, with weights published on Hugging Face.
How did it score on Artificial Analysis’s Intelligence Index?
Artificial Analysis reported a score of 38.9 on Intelligence Index v4.3.2, rounded to 39, after testing at its default xhigh reasoning effort.
Is Muse Spark Open free to use?
The weights are available without charge, but the supplied source says the model’s license has conditions. Paid API access was listed at a median of $1.25 per million input tokens and $4.25 per million output tokens across four providers.
What tasks showed particular strengths?
Artificial Analysis reported the top AA-Omniscience score among 167 open-weight entries and the top MMMU-Pro visual-reasoning score among 66 tested entries. It also reported a 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models in its comparison.
How did it perform on terminal-agent tasks?
Meta reported 76.4% on Terminal-Bench 2.1, while Artificial Analysis measured 8% on the harder Terminal-Bench 4.0 benchmark. The supplied results do not establish how it will perform on every coding or agent workflow.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
