Muse Glimmer Release: Meta Superintelligence Labs' 30B Agentic Multimodal Model

Meta Superintelligence Labs has released Muse Glimmer, a 30B multimodal model specifically designed for local agent workloads. This model is available on Ollama under the Apache 2.0 license, providing a high-context, open-weight alternative for coding agents and personal assistants.

Local Deployment for Coding Agents and Assistants

Muse Glimmer is optimized to power agentic applications running locally. It supports a context length of 128K+ tokens and is compatible with several agent frameworks and personal assistant tools via the ollama launch command.

Supported integrations include:

  • Coding Agents: Claude Code, Codex, Pi, and OpenCode.
  • Personal Assistants: OpenClaw and Hermes.
  • General Integrations: GitHub Copilot and other tools listed in Ollama's integration documentation.

Controllable Reasoning Strength

Muse Glimmer allows users to adjust the reasoning strength to balance performance and speed. The model supports four distinct levels:

  • High and XHigh: Recommended for complex coding and agentic tasks.
  • Low and Medium: Recommended for scenarios where execution speed is the prioritized metric.

Apple Silicon Optimization via MLX Engine

On Apple Silicon, Muse Glimmer achieves state-of-the-art performance through Ollama's MLX engine. This implementation introduces two primary technical enhancements:

DFlash Support

Building on previous multi-token prediction (MTP) support, the MLX engine now supports DFlash. This technology enables Muse Glimmer to run 1.5–1.8× faster on Apple Silicon hardware.

Native Image Understanding

Muse Glimmer features a dedicated 1.8B-parameter perception encoder, providing native image understanding. The MLX engine's new image input support allows coding agents to perform low-latency, sequential tool calling for image-based tasks, including:

  • Converting drawings or mockups into websites and applications.
  • Powering computer-use applications via screenshots.
  • Extracting data from documents, receipts, and charts.

Sources

Related

  • Dispatch
    Meta Muse Glimmer 30B Release

    Meta has released Muse Glimmer, a 30B parameter multimodal model distilled from Muse and released under Apache 2.0, optimized for local agentic use cases like coding and document analysis.

  • Dispatch
    Meta Muse Glimmer 30B Release Notes

    Meta has released Muse Glimmer, a 30-billion-parameter open-weights model optimized for local agentic workflows, tool use, and multimodal reasoning on consumer hardware.

  • Dispatch
    Muse Spark 1.1 Release Notes

    Meta Superintelligence Labs has released Muse Spark 1.1, a multimodal reasoning model optimized for agentic tasks, coding, and computer use, now available via the new Meta Model API.

  • Dispatch
    MiniMax M2 Release on Ollama Cloud

    MiniMax M2 is a high-performance model optimized for coding and agentic workflows, now available on Ollama's cloud platform.

  • Dispatch
    Meta Muse Code beta and Muse Spark 1.2 release: capabilities, design, and community reaction

    Meta released Muse Code (beta) and the Muse Spark 1.2 model, a more capable coding agent that uses async background agents, a persistent event log, and long‑horizon training to improve code generation and kernel optimization.