TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Meta says Muse Spark Open is available today as an open-weight model for coding, multimodal use and agentic tasks. The company reports 816 billion total parameters, a one-million-token context window and support for deployment across cloud services and, with two M5 Ultra Mac Studios, on a desktop.
Meta has released the weights for Muse Spark Open, a multimodal model it says is designed for coding and agentic work, and made it available through cloud inference providers and model repositories. Meta describes it as an 816-billion-parameter model with 42 billion active parameters and a one-million-token context window; it also says the model can run on a desktop using two Apple M5 Ultra Mac Studios.
Meta says Muse Spark Open is built for long-horizon tasks, software engineering and multi-step reasoning. The company has published the weights through Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models. Hosted inference is offered by Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, according to Meta.
The model is supported by inference engines including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp and Unsloth. Meta says it has optimized the release for AMD, Apple, Intel and NVIDIA hardware and that developers can customize it using PyTorch and TorchTitan. The weights are released under the Muse Model License, which Meta says has limited commercial terms and allows many developers to use the model for free.
Meta also describes a desktop setup using two M5 Ultra Mac Studios and the MLX framework. The company says each machine can provide up to 512 GB of unified memory and that the pair can hold the model. In tests it describes as running entirely on-device, Meta says Muse Code used the model to make and build changes to the open-source iOS app Ice Cubes and to operate DaVinci Resolve, reviewing footage and cropping selected moments vertically.
Open Weights Reach Desktop Setups
The release gives developers access to model weights they can run through several inference stacks, adapt for particular uses and deploy on supported hardware. That flexibility can matter for teams weighing hosted services against running models on their own infrastructure. The one-million-token context window and the claimed desktop configuration are notable features, though practical performance will depend on hardware, software configuration and the task.
Meta’s local-use demonstration points to a broader option for users who want agentic workflows to run on their own machines. But the stated setup requires two high-end Mac Studios, so it does not establish that the model is practical on ordinary consumer desktops. The source does not provide a purchase cost or comparative performance results for this configuration.
As an affiliate, we earn on qualifying purchases.
How Meta Describes the Release
Meta presents the weights as part of its effort to make advanced models available beyond its own hosted products. This release combines downloadable weights, hosted API access and support for multiple inference engines. That distribution gives developers several routes to try the model, but access is governed by the Muse Model License; users should review its terms before commercial deployment.
The announcement includes company-reported hardware and task demonstrations, alongside safety claims. Meta says Muse Spark Open received additional safety post-training because open-weight releases carry distinct considerations. It says the model met release thresholds under the Meta Superintelligence Scaling Framework, with details in the model card and a fuller evaluation report expected later.
““The most capable US open-weight model to date.””
— Meta
high-performance desktop computer for AI modeling
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Results Still Pending
The announcement’s performance descriptions and task demonstrations come from Meta. It references key benchmarks and an evaluation methodology report, but the supplied material does not include benchmark scores or enough detail to independently compare the model with alternatives. The company says a full safety evaluation report will be available soon.
It is also unclear from the announcement how the model performs across the listed hardware and inference engines, what the two-Mac-Studio setup costs, or what latency and reliability developers should expect in everyday use. Meta says the license allows many developers to use the weights for free, but the exact commercial limits depend on the license terms.
cloud inference servers for AI models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Developers Begin Testing the Weights
Developers can access the weights through the listed model repositories and test hosted versions through Meta’s inference partners. Those evaluations can help establish how the model performs on real coding and agentic workloads, and how its requirements differ by hardware and software stack.
Meta says a full evaluation report is forthcoming. Its release should add detail on safety testing and the framework thresholds Meta says the model met. Developers considering commercial use will also need to check the Muse Model License and assess the model on their own workloads.
multimodal AI model 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?
Muse Spark Open is an open-weight, natively multimodal model that Meta says is designed for coding and agentic work. Meta reports 816 billion total parameters, 42 billion active parameters and a one-million-token context window.
Where can developers get the weights?
Meta lists Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models as download sources. Hosted inference is available through seven named providers, including Together AI and Fireworks AI.
Can it run on a desktop?
Meta says it can run locally on two Apple M5 Ultra Mac Studios using the MLX framework. The announcement does not establish that the model can run on a typical desktop computer.
Can businesses use the model commercially?
The weights use the Muse Model License. Meta says its limited commercial terms allow many developers to use the model for free; prospective users should check the license for the terms that apply to their use.
Has Meta published independent benchmark and safety results?
The supplied announcement describes Meta’s evaluations and points to a model card. Meta says a full evaluation report will be available soon; the announcement does not provide benchmark scores or independent validation.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
