Meta released Muse Glimmer, a 30B open-source AI model optimized for agentic tasks including code generation and tool use. The model shows strong performance against Gemma4-31B and Qwen3.6-27B, with particularly notable advantages in agentic benchmarks like MCP Atlas. Key for SMEs: it compresses efficiently (K-Quant-Dynamic) to run on 32GB VRAM with only 0.2% performance loss, or 17GB with 1% loss, and uses Apache 2.0 licensing for unrestricted use. Meta is partnering with hardware makers to optimize deployment.