Muse Glimmer 30B released as an on-device agentic model
Original title: Muse Glimmer 30B: An On-Device Agentic Model
The Apache 2.0 model takes images, runs at about 4-bit on one GPU, and claims to beat same-size peers.
Researchers introduce Muse Glimmer, an open-source 30B causal language model with a perception encoder for autonomous agentic work on consumer hardware. It combines multi-step reasoning, tool use and failure recovery, accepts text and images, and was trained on data from more than a hundred languages. Weights are quantized to approximately 4-bit for single-GPU inference and released under Apache 2.0. The authors say it leads its size class against Gemma4-31B and Qwen3.6-27B on tool use, planning and task completion.
Unconfirmed: State-of-the-art benchmark claims are the authors’ and are not independently verified here.
- The model outputs text and is meant to run without cloud infrastructure.
- Training data covered more than a hundred languages.
- Authors compare it with Gemma4-31B and Qwen3.6-27B.
- The release license is Apache 2.0.
From the original article
1 min read at huggingface.co
We introduce Muse Glimmer, an open-source 30B parameter causal language model with perception encoder, purpose-built for autonomous agentic tasks on consumer hardware. Muse Glimmer integrates multi-step reasoning, reliable tool use and failure recovery into a single model that runs locally on consumer devices without requiring cloud infrastructure.
By Arash Ebrahim Abtahizadeh, Muna Aghamelu, Adil Ahmed · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.