Meta Muse Glimmer 30B
Meta's 30B Apache-2.0 open model for local coding agents — runs on one consumer GPU.
About
Muse Glimmer (August 10, 2026) is Meta Superintelligence Labs' first open-weight release since Llama 4 — and its most permissively licensed open model ever, under Apache 2.0. It is a 30B dense multimodal model (with a ~1.8B ViT-G/14 vision encoder) built for always-on local agent work: local coding agents, function calling, and LLM-as-a-judge evaluation, with a 131K-token context window. Quantized 4-bit builds fit under 20GB, running on a single 24GB consumer GPU or Apple Silicon Mac, with day-zero support across Ollama, LM Studio, llama.cpp, vLLM, SGLang, MLX, and ExecuTorch. Secondary reporting describes it as distilled from Muse Spark 1.2. The honest caveat, flagged by independent trackers: no independent benchmark figures (SWE-bench, OSWorld) have been published — all capability claims are Meta's own, so treat performance as unverified until third parties test it.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No independent benchmark figures published in sources read — capability claims are Meta's own
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Aug 10, 2026 release verified across Tech-Insider, NYT/The Register via secondary reporting, and HF (meta-models/Muse-Glimmer-30B); 30B dense, 131K context, Apache 2.0, local-first agent positioning consistent across sources; No independent benchmark figures published — entry explicitly flags that all capability claims are Meta's own.