Meta Superintelligence Labs 首个开源模型 Muse Glimmer 上线 Ollama
Muse Glimmer from Meta Superintelligence Labs is now available · Ollama Blog
Meta Superintelligence Labs 的首个开源模型 Muse Glimmer 已在 Ollama 上线，这是一个 30B 多模态模型，面向本地运行的智能体工作负载，上下文长度 128K+，采用 Apache 2.0 许可。它可用于 Claude Code、Codex、Pi 等编码智能体，以及 OpenClaw、Hermes 等长期运行的个人助手，并支持 low、medium、high、xhigh 四档可控推理强度。在 Apple Silicon 上，Ollama 的 MLX 引擎新增 DFlash 与图像输入支持，使 Muse Glimmer 运行速度提升 1.5×–1.8×，其 1.8B 参数感知编码器带来原生图像理解能力。
推荐理由
AI 评分 75
Meta Superintelligence Labs 首个开源模型在 Ollama 上线，读者可了解其本地智能体定位与 Apple Silicon 上的推理加速。
主题
正文
当前显示原文，中文全文译文尚未就绪。
August 10, 2026
Muse Glimmer, Meta’s newest open model and the first released by Meta Superintelligence Labs, is now available on Ollama. It’s a 30B multimodal model purpose-built for agent workloads that run locally with a 128K+ context length, released under the Apache 2.0 license.
With Ollama, you can now use Muse Glimmer to power coding agent applications such as Claude Code, Codex, Pi and more, as well as long-running personal assistants such as OpenClaw and Hermes.
On Apple Silicon, it is available via Ollama’s MLX engine, which provides state-of-the-art performance with new DFlash and image input support.
To run Muse Glimmer, download the latest release of Ollama and run:
For state of the art performance using Ollama’s MLX engine on Apple Silicon, use:
Power coding agents locally
To run Muse Glimmer with Claude Code, download Ollama and run:
For a lighter-weight coding agent, try Pi:
For personal assistant frameworks such as OpenClaw and Hermes, use:
Muse Glimmer also supports controllable reasoning strength: low, medium, high, and xhigh. Use high or xhigh for complex coding and agentic tasks, and lower strengths when speed matters more.
State-of-the-art performance on Apple Silicon with new DFlash and image input support
Ollama’s MLX engine now supports DFlash, building on previous multi-token prediction (MTP) support. With DFlash, Muse Glimmer runs 1.5×–1.8× faster on Apple Silicon.
To run Muse Glimmer using Ollama’s MLX engine, use the muse-glimmer:30b-mlx model.
Muse Glimmer’s dedicated 1.8B-parameter perception encoder gives it native image understanding. With new image input support in the MLX engine, coding agents can use it for low-latency, back-to-back tool calling on tasks that require images, such as:
Building websites or applications from a drawing or mockup; Computer use applications powered by screenshots; Reading documents, receipts, and charts
More information
Meta’s announcement; Model information