Meta Superintelligence Labs 开源 30B 端侧智能体模型 Muse Glimmer
Meta 开源了能塞进一张 RTX 5090 的 30B 智能体模型，工具调用和长程任务跑分压过同级 Qwen 和 Gemma，做本地 Agent 的可以试试。
Meta Superintelligence Labs 发布开源 30B 模型 Muse Glimmer，从前沿模型 Muse Spark 蒸馏而来，可在 32 GB 显存内运行，配合 4-bit 量化和投机解码头只需一张 RTX 5090。在智能体基准 MCPAtlas 上得 75.5 分，高于 Gemma4-31B 的 54.2 分和 Qwen3.6-27B 的 62.5 分；在 SWE-Bench Pro 上得 51.2 分。Qwen3.6-27B 在 OSWorld-Verified 和 TerminalBench 2.1 上仍然领先，说明 Muse Glimmer 的优势集中在工具调用和长程任务。模型支持文本和图像输入，覆盖一百多种语言，采用 Apache 2.0 许可证。
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://t.co/msBemXXG3j
Chat with Paper: https://t.co/QEJMqOOtnr
- arXiv cs.LG10-19 17:27原文
