The model is not simply another open alternative to a hosted chatbot. With 816 billion total parameters, 42 billion active parameters and a context window of up to 1 million tokens, Muse Spark Open is aimed at workloads where an AI system must retain extensive project history, inspect large codebases and complete multiple steps without constant human direction.
Meta announced the release as the “most capable US open-weight model to date,” positioning it for coding, agentic workflows, long-horizon reasoning and complex software engineering. The announcement is a claim about competitive standing, not an independently established ranking, but the model’s scale shows where Meta believes the open-model race is heading.
A model built for extended workflows
The technical specifications matter because agentic software development places different demands on a model than a short question and answer exchange. An agent working on a large application may need to track requirements, examine many files, plan changes, run tests, diagnose failures and revise its approach. A larger context window gives the system more room to keep those materials available during a single task.
Meta’s technical documentation identifies Muse Spark Open as a sparse mixture-of-experts model. It lists 816 billion total parameters, 42 billion active parameters and a 1,048,576-token context window. In a mixture-of-experts architecture, only part of the model is activated for a given input, which helps explain how a system can have a very large total parameter count while using a smaller active portion for each operation.
That design does not eliminate the infrastructure challenge. Meta’s documentation says the full weights still need to be loaded for serving, even though only a subset of parameters is active for individual inputs. As a result, the model may be open in terms of access to its weights, while remaining demanding to operate at scale.
Open weights, not just open access
The distinction is strategically important. A hosted model can be used through an application programming interface, but the provider controls the infrastructure, updates and often the handling of data. An open-weight model can give developers more control over deployment, customization and data flows, although those benefits depend on whether an organization can afford the hardware and engineering needed to run it.
Meta’s model collection on Hugging Face provides downloadable Muse Spark Open checkpoints and related artifacts. Meta’s documentation also says the model can be self-hosted or accessed through hosted providers. This gives companies two paths: operate the system themselves for greater control, or use an external service without building the entire serving stack.
The release therefore increases pressure on closed-model providers in a specific area. Companies developing coding agents may no longer need to choose only between expensive proprietary systems and smaller open models. A model with this parameter count and context capacity could support more ambitious experimentation, particularly for organizations working with sensitive source code or specialized internal tools.
Evidence remains incomplete
The main uncertainty is how the model performs outside Meta’s own presentation. The announcement includes details about availability, safety evaluations and the Muse Model License, but the supplied material does not establish how those evaluations compare with those of competing systems.
Baseten described the launch as competitive with frontier systems, while raising a significant qualification about Meta’s comparisons. According to Baseten, competing models were assessed through each laboratory’s own coding agent and tooling rather than a shared evaluation harness. That makes headline comparisons harder to interpret because software, prompts and orchestration can affect results alongside the underlying model.
The sources supplied for this article contain no opposing assessment of Muse Spark Open. That means the strongest available criticism concerns benchmark comparability, not a demonstrated failure of the model.
The next test will be practical adoption. Developers will need to determine whether Muse Spark Open’s long context and agentic coding capabilities justify its operating requirements, licensing conditions and deployment complexity. If they do, Meta’s release could help make very large open-weight models a realistic foundation for software agents. If they do not, its impressive specifications may remain more important as a signal of direction than as evidence of an immediate market leader.
Alex Carter is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Alex Carter's usual length and in Alex Carter's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.