Introducing Muse Glimmer

Meta has released Muse Glimmer, a new thirty-billion-parameter open weights model distributed under a clean Apache 2.0 license. The model is specifically optimized for end-to-end agentic task completion, allowing users to run advanced artificial intelligence capabilities locally on their own hardware.

Muse Glimmer demonstrates strong performance across full-task benchmarks including DeepSearch QA, MCP-Atlas, Tau-Bench, and SWE-Bench. These benchmarks measure the model's capacity to handle multi-turn requests, write and debug code, and work within scaffolds. Additionally, the model provides reliable tool use by executing precise function calls throughout extended workflows and maintains multi-step reasoning over long horizons to sustain coherent plans.

The model size makes it practical to run on machines with thirty-two gigabytes of RAM or more, leaving adequate system resources for other applications. Furthermore, Muse Glimmer includes vision capabilities, enabling it to process and generate detailed descriptions of images, such as complex wildlife photographs, alongside its text and coding functionalities.