Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

Muse Glimmer：专为全天候本地智能体工作流优化的 300 亿参数模型

Today, we’re introducing Muse Glimmer, the next model from Meta Superintelligence Labs, and open sourcing the model weights under a permissive Apache 2.0 license. Muse Glimmer is a 30-billion-parameter model optimized for always-on local agent workflows. It’s small enough to run on a Mac or PC with a single consumer GPU, enabling use cases that range from local agents and function calling, to local coding, and LLM-as-a-judge evaluation.

今天，我们正式推出 Meta 超智能实验室（Meta Superintelligence Labs）的最新模型 Muse Glimmer，并以宽松的 Apache 2.0 许可证开源其模型权重。Muse Glimmer 是一个拥有 300 亿参数的模型，专为全天候本地智能体工作流而优化。它的体积足够小，可以在配备单块消费级 GPU 的 Mac 或 PC 上运行，支持从本地智能体、函数调用，到本地编程以及“大模型作为裁判”（LLM-as-a-judge）评估等多种应用场景。

Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category. Foundation models have achieved remarkable capabilities across reasoning, code generation, and tool use — yet most deployments still depend on cloud infrastructure and network access. Running models locally enables you to use AI anywhere, anytime, with or without an internet connection. This is increasingly viable: the open source community has shown that smaller models, when trained effectively, can approach frontier-level performance on targeted tasks. Muse Glimmer is optimized for these local use cases.

与同等规模的领先模型相比，Muse Glimmer 在关键的智能体用例和基准测试中表现出色。基础模型在推理、代码生成和工具使用方面已经取得了显著成就，但大多数部署仍依赖于云基础设施和网络连接。在本地运行模型使您能够随时随地使用 AI，无论是否有互联网连接。这一点正变得越来越可行：开源社区已经证明，较小的模型如果经过有效训练，可以在特定任务上接近前沿水平的性能。Muse Glimmer 正是为这些本地用例而优化的。

Keeping with our long tradition of sharing fundamental AI research, we’re releasing Muse Glimmer open weights today on Hugging Face, along with developer documentation to help you start building and running your own agents. Muse Glimmer is built to work with the tools developers already use. Optimized integrations on llama.cpp, MLX, and ExecuTorch will land in the coming days, so you can go from download to working agent in minutes.

秉承我们分享基础 AI 研究的长期传统，我们今天在 Hugging Face 上发布了 Muse Glimmer 的开源权重，并附带了开发者文档，以帮助您开始构建和运行自己的智能体。Muse Glimmer 旨在与开发者现有的工具协同工作。针对 llama.cpp、MLX 和 ExecuTorch 的优化集成将在未来几天内上线，让您可以在几分钟内从下载模型过渡到运行智能体。

How We Trained Muse Glimmer

我们如何训练 Muse Glimmer

An agent that manages your schedule, drafts your messages, organizes your files, and learns how you work needs deep access to personal context. It also needs several capabilities working in concert: long-horizon execution, precise tool calling, multimodal understanding, long-context memory, and instruction following. We designed Muse Glimmer to balance capability against the memory and compute constraints of local hardware. This required a compact architecture, a novel distillation recipe that transfers agentic reasoning from a much larger teacher model, and inference optimizations — including quantization — to meet latency expectations.

一个能够管理您的日程、起草消息、整理文件并学习您工作方式的智能体，需要深入访问个人上下文。它还需要多种能力协同工作：长周期执行、精确的工具调用、多模态理解、长上下文记忆以及指令遵循。我们在设计 Muse Glimmer 时，旨在平衡能力与本地硬件的内存及计算限制。这需要紧凑的架构、一种将智能体推理能力从更大的教师模型迁移过来的新型蒸馏方案，以及包括量化在内的推理优化，以满足延迟预期。

We achieved this in the following phases: 我们通过以下阶段实现了这一目标：

Pre-Training. We trained Muse Glimmer on Muse Spark’s outputs using logit distillation, leveraging a similar data mix as the teacher.
预训练。 我们使用 Logit 蒸馏技术，利用与教师模型相似的数据组合，在 Muse Spark 的输出上训练了 Muse Glimmer。
Mid-Training. We trained the model on longer-context, more agent-heavy data with richer reasoning traces, alongside organic data.
中期训练。 我们在更长上下文、更侧重智能体的数据（包含更丰富的推理轨迹）以及有机数据上对模型进行了训练。
Post-Training. We combined supervised fine-tuning with a mix of on-policy distillation and reinforcement learning across general, reasoning, coding, and agentic domains. Muse Glimmer was evaluated under the standards set out in Meta’s Advanced AI Scaling Framework and assessed for open-weight release across all relevant categories.
后训练。 我们将监督微调与策略内蒸馏（on-policy distillation）及强化学习相结合，涵盖了通用、推理、编码和智能体领域。Muse Glimmer 按照 Meta 高级 AI 扩展框架（Advanced AI Scaling Framework）设定的标准进行了评估，并在所有相关类别中通过了开源权重发布的评估。

Built for Agents: What Muse Glimmer Can Do

为智能体而生：Muse Glimmer 的能力

Building effective agents requires key capabilities working together to achieve the user’s goals. Muse Glimmer is trained and evaluated across each of the following: 构建高效的智能体需要多种关键能力协同工作，以实现用户的目标。Muse Glimmer 在以下每个方面都经过了训练和评估：

End-to-end Agentic Task Completion. Muse Glimmer achieves strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, 𝛕-Bench and SWE-Bench, which measure its ability to work within scaffolds, write and debug code, and resolve multi-turn requests from start to finish.
端到端智能体任务完成。 Muse Glimmer 在包括 DeepSearch QA、MCP-Atlas、𝛕-Bench 和 SWE-Bench 在内的全任务基准测试中取得了很高的成功率，这些测试衡量了它在框架内工作、编写和调试代码以及从头到尾解决多轮请求的能力。
Reliable Tool Use. The model handles a wide range of function calls, invoking tools with precise schemas throughout extended workflows.
可靠的工具使用。 该模型能够处理广泛的函数调用，在扩展的工作流中通过精确的模式调用工具。
Multi-Step Reasoning. Muse Glimmer chains reasoning over long horizons, sustaining coherent plans across complex, extended workflows.
多步推理。 Muse Glimmer 能够进行长周期的推理链式反应，在复杂、扩展的工作流中维持连贯的计划。
Failure Recovery. When a tool call fails or returns an unexpected result, the model is trained to diagnose the error and retry rather than halt.
故障恢复。 当工具调用失败或返回意外结果时，模型经过训练可以诊断错误并重试，而不是直接停止。
Multimodal Input and Reasoning. Through a dedicated perception encoder, the model accepts interleaved text and images. This enables agents to interpret screenshots, charts, and documents alongside conversation.
多模态输入与推理。 通过专门的感知编码器，模型可以接受交错的文本和图像。这使得智能体能够在对话的同时解读截图、图表和文档。
Scaffold Compatibility. Muse Glimmer works across OpenClaw and other agentic orchestration patterns.
框架兼容性。 Muse Glimmer 可与 OpenClaw 及其他智能体编排模式兼容。
Controllable Effort. Muse Glimmer supports different reasoning strengths to select the right balance between quality and speed.
可控的计算投入。 Muse Glimmer 支持不同的推理强度，以便在质量和速度之间选择合适的平衡点。
Multilingual. Muse Glimmer is trained on data from more than 100 languages.
多语言支持。 Muse Glimmer 在来自 100 多种语言的数据上进行了训练。

Performance

性能表现

We evaluated Muse Glimmer across a broad range of benchmarks to assess the diverse capabilities required for effective autonomous agent behavior. Compared with Gemma4-31B and Qwen3.6-27B, Muse Glimmer performs strongly for its size class on several widely used LLM benchmarks. For more detail about our evaluations, see our report. 我们通过广泛的基准测试评估了 Muse Glimmer，以评估有效自主智能体行为所需的多样化能力。与 Gemma4-31B 和 Qwen3.6-27B 相比，Muse Glimmer 在多个广泛使用的大模型基准测试中，在其规模级别上表现强劲。有关我们评估的更多详细信息，请参阅我们的报告。

Optimized for Local Deployments

针对本地部署进行优化

A local agent is truly useful if it’s fast enough to feel responsive. An agent that takes minutes to reply or plan its next step breaks the flow of real work. We applied two optimizations to make Muse Glimmer run at practical speeds on consumer hardware without sacrificing quality. 只有速度足够快、响应足够灵敏，本地智能体才真正有用。如果一个智能体需要几分钟才能回复或规划下一步，就会打断实际工作的流程。我们应用了两种优化方案，使 Muse Glimmer 能够在不牺牲质量的前提下，在消费级硬件上以实用的速度运行。

Fitting the Model on Your Device. At full precision, a 30-billion parameter model would require over 55 GB of memory — far more than any consumer GPU offers. We use quantization techniques to compress the model’s weights to approximately 4-bit precision, shrinking the language model to under 20 GB. This leaves enough headroom for the model’s working memory (its “KV cache”), the perception encoder for image understanding, and the speculative decoding drafter to run simultaneously within a 24 GB or 32 GB envelope. We validated that this compression introduces minimal to no degradation on agentic tasks.
适配您的设备。 在全精度下，一个 300 亿参数的模型需要超过 55 GB 的内存，这远超任何消费级 GPU 的容量。我们使用量化技术将模型权重压缩到约 4-bit 精度，将语言模型缩小到 20 GB 以下。这为模型的运行内存（KV 缓存）、用于图像理解的感知编码器以及推测解码草稿模型留出了足够的空间，使其能够同时在 24 GB 或 32 GB 的显存限制内运行。我们验证了这种压缩对智能体任务的影响微乎其微，甚至没有退化。
Faster Generation Through Speculative Decoding. Language models normally generate text one token at a time, which can feel slow during long reasoning chains or multi-step tool calls. Muse Glimmer ships with a lightweight “drafter” model based on DFlash — a small companion network that proposes entire blocks of tokens at once. The main model then verifies these proposals in parallel, accepting correct tokens and correcting wrong ones. This technique lets Muse Glimmer…
通过推测解码实现更快的生成。 语言模型通常一次生成一个 Token，这在长推理链或多步工具调用中会显得缓慢。Muse Glimmer 附带了一个基于 DFlash 的轻量级“草稿”模型——这是一个小型配套网络，可以一次性提出整个 Token 块。主模型随后并行验证这些建议，接受正确的 Token 并纠正错误的 Token。这项技术让 Muse Glimmer……