Meta Muse Glimmer brings local AI agents to consumer GPUs
| Source: AI News (ainews.com)
Tags: Meta, Muse Glimmer, Apache 2.0, local AI, consumer GPU, AI agents, HuggingFace, open-source
Meta released Muse Glimmer, a 30B-parameter model under Apache 2.0 for local AI agents on consumer GPUs — leading competitors on MCP Atlas (75.5 vs. 54.2 for Gemma4-31B), SWE-Bench Pro (51.2 vs. 36.9), and DeepSearch QA (74.6 vs. 61.7), with weights on HuggingFace for local coding, function calling, and on-device personal agents.
Details
Meta's Superintelligence Labs released Muse Glimmer, a 30-billion-parameter model available under Apache 2.0 on HuggingFace. The model is designed to run locally on consumer GPUs and targets agentic use cases: local coding agents, function calling, LLM-as-a-judge evaluation, and personal agents with access to private context like schedules, messages, and files — with no cloud dependency. On benchmarks, Muse Glimmer leads several agent-focused evaluations against size-comparable models. On MCP Atlas it scored 75.5 versus Gemma4-31B's 54.2 and Qwen3.6-27B's 62.5 — a 21-point gap over Gemma. On SWE-Bench Pro (complex software engineering), it scored 51.2 versus 36.9 and 50.2. On DeepSearch QA, it scored 74.6 versus 61.7 and 71.1. Qwen3.6-27B outperforms Muse Glimmer on several other benchmarks: GDPval-AA (1,141 vs. 953), OSWorld-Verified (75.6 vs. 65.9), and TerminalBench 2.1 (60.7 vs. 51.7). The competitive picture is strong but not a clean sweep across all agent tasks. The Apache 2.0 license removes commercial use barriers. Running locally means no per-token cost and the ability to connect agents to truly private organizational data. For enterprise teams evaluating on-premise AI, this is a credible option at the 30B scale.