Meta opens Muse Glimmer under Apache 2.0 — a 30B model that runs on one consumer GPU
Meta Superintelligence Labs released the weights for Muse Glimmer, a 30-billion-parameter dense model tuned for local agentic work, on Hugging Face under Apache 2.0. Meta's benchmarks put it ahead of Gemma4-31B and Qwen 3.6 27B. Zuckerberg says the weights for Muse Spark 1.2, the company's latest foundation model, follow in the coming weeks.
The licence is the story. Apache 2.0 is not the Llama community licence with its acceptable-use appendix and its 700-million-user tripwire — it is the same permissive grant that Z.ai used for GLM-5.2 and DeepSeek used for V4 Flash. Meta spent two years arguing that its bespoke licence was open enough. It has now stopped arguing and adopted the terms its competitors already offered.
The engineering target is equally specific. Thirty billion parameters, dense rather than mixture-of-experts, 131K context, support for more than 100 languages, and a 4-bit quantisation that compresses the whole thing under 20GB — which is to say, onto a single consumer card, at a claimed 3.1x speedup. Meta trained it by distilling Muse Spark outputs and then refining with reinforcement learning, and pointed it at tool use, coding, and LLM-as-judge work rather than chat.
That combination — permissive licence, agent-shaped tuning, fits on hardware a developer already owns — is the configuration that has been taking download share from American labs all year. MiniMax open-sourcing H3 and NVIDIA's Alpamayo release the same week make the pattern hard to miss.
What to watch is whether Muse Spark 1.2 actually ships open. A 30B distillation is a real contribution and also a low-risk one; the foundation model is the commitment that would cost Meta something. The company has promised it without naming a date or a licence, and it promised it five days after launching the same model closed.
VentureBeat — Meta returns to open source with Muse Glimmer, an Apache 2.0 licensed 30B parameter AI model optimized for agents → · Constellation Research — Meta releases open weight Muse Glimmer model with open Muse Spark 1.2 on tap → · CNBC — Meta launches Muse Glimmer open-weight AI model →