Meta’s new Muse Glimmer model is now available in Ollama, giving developers a way to run a 30-billion-parameter multimodal coding model locally instead of relying only on hosted systems. The model is the first open release from Meta Superintelligence Labs and is distributed under the Apache 2.0 license.
Ollama says the model is designed for local coding agents and can work with image inputs, which means a developer can pair text instructions with screenshots or visual context during a coding workflow. The release also uses Ollama’s MLX engine with native DFlash support, aimed at improving performance on Apple silicon.
The practical value is control: teams can experiment with agentic coding workflows on their own hardware, with fewer data-sharing and latency concerns. The tradeoff is that a 30B model still requires capable local equipment, so this is most useful for developers with modern machines rather than casual users on low-end laptops.