Meta Superintelligence Labs 发布开源 30B 端侧 Agent 模型 Muse Glimmer，从自家前沿模型 Muse Spark 蒸馏而来，Apache 2.0 协议，支持文本和图像输入、百余种语言，4-bit 量化加投机解码头后可在单张 RTX 5090、32 GB 显存内运行。
原文给出与 Gemma4-31B、Qwen3.6-27B 的具体基准对比和硬件门槛，构建本地 Agent 的读者可据此评估是否采用。
Banger paper from Meta Superintelligence Labs on on-device agent models.
(bookmark it)
If you build agents that need to run locally, this one is worth your time.
Muse Glimmer is an open 30B model, distilled from Meta's frontier model Muse Spark and designed to fit in 32 GB of VRAM. With 4-bit quantization and a speculative decoding head, it runs on a single RTX 5090.
On MCPAtlas, it scores 75.5, against 54.2 for Gemma4-31B and 62.5 for Qwen3.6-27B. On SWE-Bench Pro, it scores 51.2, against 36.9 and 50.2.
Qwen3.6-27B still leads on OSWorld-Verified and TerminalBench 2.1, so the gains are concentrated in tool use and long-horizon tasks.
The model takes text and images, covers more than a hundred languages, and is released under Apache 2.0.
https://arxiv.org/abs/2610.24628
Chat with Paper: https://academy.dair.ai/papers/muse-glimmer-30b-an-on-device-agentic-model-2610.24628
来源：DAIR.AI · x.com