Meta releases Muse Glimmer for local AI agents
Meta has released Muse Glimmer, a 30-billion-parameter model designed for local AI agents running on consumer hardware. Superintelligence Labs published its weights on Hugging Face under the Apache 2.0 licence. Meta says developers can use the model for local coding, function calling, agent workflows and LLM-as-a-judge evaluation, including applications that handle private schedules, messages and files without relying on a cloud-hosted model.
In Meta’s comparisons, Muse Glimmer led Gemma4-31B and Qwen3.6-27B on five of eight general-agentic benchmarks. It scored 75.5 on MCP Atlas, 74.6 on DeepSearch QA and 47.6 on WildClawBench. Qwen3.6-27B led several other evaluations, including GDPval-AA, SkillsBench and OSWorld-Verified. Coding results were mixed: Muse Glimmer led SWE-Bench Pro with 51.2, while Qwen3.6-27B performed better on SWE-Bench Verified and TerminalBench 2.1.
Meta says approximately 4-bit quantisation reduces the model’s memory use to under 20 GB, compared with more than 55 GB for full precision. The company targets systems with 24 GB or 32 GB of memory and tested a quantised version on MacBook M4-Max, MacBook M5-Max and RTX-5090 hardware. Integrations with llama.cpp, MLX and ExecuTorch are expected in the coming days.
Written by Sakhanda Wire AI from reporting by AI News. Facts, quotes and full context are in the original article below.