Meta launched a new open-weight AI model, Muse Spark Open, on Thursday, the tech giant announced in a blog post. Muse Spark Open is a coding-focussed model that Meta says can also work with screenshots, video clips, and documents in real time.

The company said it is made for complex software tasks and has strong safeguards, though it did not say what they are.

What are open-weight models?

An open-weight model is an AI model whose core components — the numerical parameters learned during its training, known as its weights — are released publicly. Anyone can download the model, run it on their computer, study how it works, and adapt it to their needs, subject to the model’s licence and usage terms.

Also read: What are open weight AI models and why are they dividing Big Tech?

What is Meta Muse Spark Open

The coding-focussed AI model is built for ‘‘coding and agentic work, long-horizon tasks, complex software engineering, and multi-step reasoning’’, the company claims. Meta said its visual reasoning runs in a real working environment, not through fixed, pre-written steps.

‘‘Muse Spark Open is derived from the Muse Spark model we released in August, with additional safety post-training to make it more robust against misuse as a standalone model without system guardrails’’, Meta said.

'Most Capable’

“Muse Spark Open is the most capable US open weight model to date,” Meta’s chief of AI Alexandr Wang wrote on X.

Meta claims Muse Spark Open is the first model ‘‘in its class’’ to run on a desktop, using two

Apple M5 Ultra Mac Studios with the latest MLX framework to build a “supercomputer” on a user’s desk.

Availability

The model is available through API platforms including Baseten, Fireworks AI, Groq, and others. Its weights can be downloaded from Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face, and Microsoft Foundry Models.

It runs on inference engines such as SGLang, vLLM, Hugging Face Transformers, and Apple MLX, with optimisations for common developer tools and chips from Apple, Intel, and Nvidia AMD. Developers can customise it using PyTorch, a Meta developed machine learning framework.

Meta said it is releasing the weights under the Muse Model License, which has limited commercial terms that allow many developers to use the model for free.