Abstract
Muse Glimmer is a 30B-parameter open-source causal language model with a perception encoder that performs multi-step reasoning and tool use locally on consumer hardware via 4-bit quantization.
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. The model accepts text and image inputs and produces text outputs, was trained on data from over a hundred languages, and is quantized to approximately 4-bit precision for efficient single-GPU inference. Muse Glimmer achieves state-of-the-art performance for its size class on agentic benchmarks, setting a new standard compared to the models of its size: Gemma4-31B and Qwen3.6-27B on tool use, multi-step planning, and agentic task completion. We release the model under the Apache 2.0 license to foster open-source development of capable, safe, local AI agents.
Community
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- IronLLM: Forging Compact Edge-Native Language Models for Real-Time Embodied Intelligence (2026)
- ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search (2026)
- Instella-MoE Technical Report (2026)
- MintAct: A Unified Visual Agent for Digital Environments (2026)
- LoongSage: A Production-Grade, High-Efficiency Agentic RL Framework for Frontier LLMs with Validated Recipes (2026)
- Hunyuan-A13B Technical Report (2026)
- AutoDataBench: A Data-centric Testbed for Accelerating Auto Research (2026)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
The abstract says the model is quantized to about 4 bits for single-GPU inference. Were the agentic benchmark scores in the paper measured on that 4-bit build, or on the full-precision weights? For tool use over long multi-step runs, small quantization errors can add up in ways a single-turn benchmark does not show. I would also like to know if anyone on the team tried the Q4_K_M GGUF with CPU-only llama.cpp on a desktop with 32 GB of RAM, and whether the DFlash drafter still helps when there is no GPU.
Get this paper in your agent:
hf papers read 2610.24628 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 3
Datasets citing this paper 0
No dataset linking this paper