Home / Models / Muse Glimmer (30B)

Muse Glimmer (30B)

Meta (Meta Superintelligence Labs) · Multimodal model

Muse Glimmer is a 30-billion-parameter open-weight model Meta released on August 10, 2026 under Apache 2.0. It is derived from Muse Spark and built to run agents locally on one consumer GPU. It accepts interleaved text and images.

Current Open weights
ReleasedAug 10, 2026
Context window—
Input → outputtext, image → text
API price—
Consumer accessFree weights on Hugging Face under Apache 2.0; runs on a single consumer GPU or Mac (~4-bit build under 20 GB).

Best for

Always-on local agents, function calling and local coding on a single consumer GPU or Mac.

Overview

Muse Glimmer is a 30-billion-parameter open-weight model Meta released on August 10, 2026 under Apache 2.0. It is derived from Muse Spark and built to run agents locally on one consumer GPU. It accepts interleaved text and images.

Strengths and limits

  • Apache 2.0 open weights — Meta's most permissive license yet
  • Fits 24–32 GB memory when quantized to ~4-bit
  • Trained for multi-step reasoning, tool use and recovering from failures
  • Training data and code not released (weights only)
  • Official benchmark numbers only published as an image