← Back to Discover

Muse Glimmer

Open-weight model of roughly 30 billion parameters that takes images and text and replies in text, published on Hugging Face under the Apache 2.0 licence.

Open models

What it does

Muse Glimmer 30B is a model repository on Hugging Face under the meta-models organisation, which the page shows as Meta Inc. It is tagged image-text-to-text and conversational, so it accepts images alongside text in a chat and answers in text. The weights are stored in safetensors format in BF16 and total about 29.8 billion parameters. The repository is not gated and carries the Apache 2.0 licence, and the page gives Transformers code for loading it with a pipeline or directly. The bundled chat template describes a tool-calling protocol with a reasoning channel, tool channels and a user channel, and it can render image and video parts of a message. Reasoning strength defaults to high unless it is set otherwise. The page lists benchmark results for coding, maths and multimodal tasks, all marked as not verified. It also indicates hosted inference through third-party providers, deployment to SageMaker and Azure, and the existence of quantized versions.

Capabilities

  • Accepts image and text input and replies in text
  • Safetensors weights in BF16, about 29.8B parameters
  • Apache 2.0 licence with a repository that is not gated
  • Chat template with a tool-calling protocol and a reasoning channel
  • Loadable with Hugging Face Transformers

When to use it

  • You want an openly licensed multimodal model that you can download and run yourself
  • You are building a chat or agent setup that needs tool calls and a reasoning channel
  • You want to try a model through Transformers or a hosted inference provider