Introducing Muse Glimmer
Introducing Muse Glimmer Meta are back in the open weights game! Muse Glimmer is a brand new 30B model under a clean Apache 2.0 license (a step up from the janky Llama licenses of old). They claim to have optimized it for exactly the kind of things I'm looking for in a local model: End-to-end Agentic Task Completion. Muse Glimmer achieves strong success rates on full-task benchmarks including…
Meta has unveiled Muse Glimmer, a cutting-edge 30B model released under an Apache 2.0 license. This new model boasts impressive capabilities in end-to-end agentic task completion, excelling in benchmarks such as DeepSearch QA, MCP-Atlas, 𝛕-Bench, and SWE-Bench. Muse Glimmer demonstrates strong performance in reliable tool use, multi-step reasoning, and handling complex workflows.
The model's versatility is evident in its ability to perform various function calls, including invoking tools with precise schemas throughout extended tasks. A demonstration of Muse Glimmer's prowess is showcased by a user-generated image, where the model accurately describes a rocky shoreline, two pelicans, and surrounding birds. The detailed description highlights the pelicans' distinctive features and plumage, as well as the presence of smaller birds and the overall atmosphere of the scene.
Muse Glimmer's reputation as a reliable vision model is further reinforced by its ability to process and describe images with remarkable accuracy. This makes it an ideal choice for users with 32 GB of RAM or more, as it leaves ample room for running other applications simultaneously. The release of Muse Glimmer marks Meta's continued commitment to open weights and advancing the capabilities of local models.
Written by urgent.news from Simon Willison's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
