🔍 Read the full analysis: Muse Spark Open Faces A Key Question: Is Its Price Worth It? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 weights on Hugging Face and a 38.9 score on Artificial Analysis’s Intelligence Index. Its strongest reported results include low hallucination rates and visual reasoning, but its API list prices match the higher-scoring Muse Spark 1.3, while several open competitors cost less.
Meta released Muse Spark Open on October 15, making the 816-billion-parameter model available on Hugging Face. It scored 38.9 on Artificial Analysis’s Intelligence Index, but its listed API price matches the closed Muse Spark 1.3, which scored 45.1 on the same index and reasoning setting. The gap puts the model’s performance, cost and license conditions at the center of the decision for developers considering it.
Muse Spark Open is a mixture-of-experts model with 42 billion active parameters per token and 256 experts, eight active for each token. Meta lists a one-million-token context window and support for text, images, speech and audio, and video as inputs; it generates text. The weights are available in BF16 and NVFP4 formats.
Meta’s model card estimates self-hosting requires 700 GB of GPU memory for NVFP4 or 2 TB for BF16. It also says the model is optimized to run on two Mac Studios with M5 Ultra using MLX. Those requirements make hosted access a practical route for many teams, but that access is not free: Artificial Analysis reported a median provider list price of $1.25 per million input tokens, $0.15 per million cached input tokens and $4.25 per million output tokens across four third-party providers.
Artificial Analysis tested the model at its default xhigh reasoning effort on release day. The score of 38.9 is the highest reported for an open-weight model from a US lab in the source’s comparison, and tenth among open-weight models overall. Nine models from Chinese labs scored higher. The index and rankings reflect Artificial Analysis’s evaluations, rather than results independently established here.
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 a 38.9 Score
The pricing comparison changes the value proposition. Artificial Analysis calculates a weighted cost of $0.84 per Intelligence Index task for Muse Spark Open. In that comparison, Xiaomi’s MIT-licensed MiMo-V2.6-Pro scores 46.3 for $0.13 per task, while DeepSeek V4.1 Flash scores 39.5 for $0.27. MiMo-V2.6-Flash scores 37.9 for $0.06. These figures use the source’s index-task cost estimate; actual bills will depend on usage and provider pricing.
Among models with their own restrictive licenses, Muse Spark Open’s estimated task cost is lower than Step 5 Preview, Kimi K3 and GLM-5.3 in the cited comparison. But developers weighing it against MIT-licensed alternatives may find the price difficult to justify on the headline score alone. The weights are free to download, but self-hosting carries substantial memory requirements, and using a provider brings per-token charges.
The model’s results are uneven across tasks. Artificial Analysis reported 24% hallucination rate, the lowest among the 15 highest-scoring open-weight models it tested, as well as first place among the open-weight entries on its AA-Omniscience and MMMU-Pro evaluations. Yet on Terminal-Bench 4.0, it scored 8%, compared with 33% for Muse Spark 1.3, 35% for MiMo-V2.6-Pro and 42% for GLM-5.3. For teams, the choice may depend more on workload than on one overall score.
high performance GPU for AI training
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Meta’s Open Model Gap
Muse Spark Open is the first open-weight model in Meta’s flagship Muse Spark line, according to the source. Meta’s earlier open models were smaller: Muse Glimmer, released in August 2026 with an Apache 2.0 license, scored 17 on the current index, while Llama 4 Maverick scored 9. NVIDIA’s Nemotron 3.5 Super VL, released the same day as Muse Spark Open, scored 26.5 in the cited evaluation.
The new release changes the scale of Meta’s open-weight offering, but it does not lead the overall open-weight field on the reported score. The source lists MiMo-V2.6-Pro at 46.3 and several other models above Muse Spark Open. Within Meta’s own lineup, Muse Spark 1.3 scored 45.1 at the same xhigh effort and list price, while Muse Wonder scored 52.4. These comparisons are useful reference points, though benchmark scores do not establish how a model will perform on every application.
As an affiliate, we earn on qualifying purchases.
Limits of the Current Comparison
The reported figures do not settle whether Muse Spark Open is a good fit for a particular product. Benchmark results vary by task, and the source does not provide evidence here about performance across developers’ own workloads, operating costs at their usage levels or the effect of provider-specific implementation choices. The four-provider median is a snapshot of listed rates, not a guarantee of the price a buyer will pay.
The available source material also does not spell out the license conditions in detail, so teams should review the model’s license before deployment. Meta’s memory estimates describe self-hosting requirements, but the material does not establish hardware availability or total operating costs for a given organization. The relative value could differ for users who prioritize visual reasoning, long documents or lower hallucination rates over agent performance or price.
As an affiliate, we earn on qualifying purchases.
What Developers Should Check
Developers considering the model can compare its license terms and hosting options with their intended use, then test representative tasks before committing. For hosted access, they can check current provider rates and estimate costs using their expected input, cached input and output volumes. Those details matter because input accounted for about 80% of the cited $0.84 task cost: $0.67, including $0.40 for cache reads, $0.25 for cache writes and $0.02 for uncached input; output accounted for $0.17.
Further evaluations could clarify how the model performs on practical workloads and whether the initial benchmark pattern holds. For now, the reported evidence points to a capable open-weight model with notable results in some areas, a weaker showing on one harder terminal-agent benchmark and API rates that match a higher-scoring closed Meta model. Its value will depend on the task, license and serving choice.
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 available on Hugging Face.
How did Muse Spark Open score?
Artificial Analysis reported a score of 38.9 on its Intelligence Index v4.3.2, testing at its default xhigh reasoning effort.
Is the model free to use?
The weights are available to download, but the source does not detail all license conditions. Hosted API access has listed per-token charges, and self-hosting requires substantial GPU memory according to Meta’s model card.
How does it compare with Muse Spark 1.3?
In the cited comparison, Muse Spark 1.3 scored 45.1 versus 38.9 for Muse Spark Open at the same xhigh effort. Their listed API prices are the same.
Which tasks appear to suit Muse Spark Open?
The reported evaluations show strong results in visual reasoning and low hallucination rates, as well as a one-million-token context window. Its 8% score on Terminal-Bench 4.0 indicates weaker performance on that terminal-agent evaluation than several compared models.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
