Meta выпускает открытую модель Muse Glimmer: разворот после ставки на закрытый Muse Spark 1

Meta has released an open-weight model, Muse Glimmer, making a sharp U-turn after betting on closed products — in particular, the Muse Spark 1 model. According to AI Business, the shift is driven by corporate market pressure: businesses demand the ability to deploy AI on their own infrastructure without sending data to the vendor's servers.

Why did Meta open the weights of Muse Glimmer?

Muse Glimmer emerged as a direct response to enterprise client requests. The corporate segment — financial institutions, medical companies, government structures — fundamentally cannot send sensitive data to an external provider's servers, even if that provider is Meta. Regulatory requirements, internal security policies, and insurance risks make working through a cloud API physically impossible for part of the market.

The closed model Muse Spark 1 gave Meta full control over the AI stack: from training to inference, from pricing to terms of use. However, this approach, according to AI Business, cut off an entire layer of potential corporate clients who have no technical or legal ability to work through a third-party cloud API. The open weights of Muse Glimmer remove this barrier — a company receives the model and deploys it wherever it sees fit.

"This shift addresses enterprise demand for stricter data control and local infrastructure systems," states the AI

Business article describing Meta's pivot.

What does business get with an open model?

Open weights are not simply a "free download." An organization receives the model files and deploys them wherever it sees fit: in its own data center, in a cloud of its choosing, or on a hybrid infrastructure. Data does not leave the company's secure perimeter, and traffic does not pass through Meta's servers.

For enterprise clients, this opens up possibilities fundamentally unavailable when working with a closed API:

• Full data control — no traffic through Meta's infrastructure
• Fine-tuning for company tasks — fine-tuning on internal datasets without transferring them to the provider
• Predictable scaling costs — without dependence on API pricing and conditions that could change at any moment
• Independent security audit — the internal team checks and validates the model independently

How Meta's pivot fits into the market

The decision to release Muse Glimmer fits into a persistent trend: major AI labs increasingly support parallel product lines — closed products with full support and ready-made APIs for those who want to get started quickly, and open weights for those who need full autonomy.

Closed models win where speed of deployment matters, minimal upfront operational costs are important, and there are no specific data transfer restrictions. Open models are indispensable in scenarios where regulatory requirements, data sensitivity, or inference scale make dependence on an external API unacceptable. The stratification of corporate demand into these two segments has apparently become clear enough for Meta to return to an open-weight strategy — even at the cost of less control over the stack.

What this means

Muse Glimmer is an acknowledgment from Meta: closed models systematically fail to reach part of the corporate market. The company is not abandoning Muse Spark 1, but is opening a second front — where the closed strategy simply does not work. For other AI labs, this is yet another signal: presence in the open-weight enterprise segment is rapidly moving from a "competitive advantage" to a basic market expectation.

*Meta is recognized as an extremist organization and is banned in Russia.