Meta releases Muse Glimmer, a 30B Apache-2.0 multimodal model built for local agentic use

Published 10 August 2026, Muse Glimmer is a dense 30B multimodal model distilled from Meta's Muse and released under Apache 2.0 — a 2B vision encoder plus a 28B text decoder, with an optional speculative-decoding drafter. It ships with day-0 support in transformers, llama.cpp, vLLM and Inference Endpoints, and is aimed squarely at privacy-sensitive local deployment: coding, document analysis and personal assistants. Published benchmarks put it ahead of Gemma4-31B and Qwen3.6-27B on agentic tasks such as MCP Atlas (75.5) and SWE-Bench Pro (51.2), while trailing Qwen3.6 on OSWorld-Verified and TerminalBench.

What it means A 30B Apache-2.0 model that runs agentic work on your own hardware puts a real floor under what you have to pay a frontier API for — and llama.cpp already ships tool-call handling for it.