Meta Superintelligence Labs has released Muse Glimmer, the first model from the organization, and it’s now available on Ollama. The model is a 30B parameter multimodal model designed for agent applications. It supports a context length of 128K+ and is released under the Apache 2.0 license.
Users can utilize Muse Glimmer to power coding agent applications such as Claude Code, Codex, Pi, and OpenCode. It also supports personal assistant frameworks like OpenClaw and Hermes. The model’s architecture includes a 1.8B-parameter perception encoder for native image understanding.
Ollama’s MLX engine provides state-of-the-art performance, particularly on Apple Silicon. The engine incorporates DFlash, a multi-token prediction (MTP) support, resulting in a 1.5×–1.8× speed increase compared to previous implementations. This allows for low-latency, back-to-back tool calling on tasks requiring image input, such as building websites from drawings or using screenshots for computer applications.
Controllable reasoning strength is supported, allowing users to adjust the model’s output based on the task. Strengths of low, medium, high, and xhigh are available. High or xhigh strengths are recommended for complex coding and agentic tasks, while lower strengths are suitable when speed is a priority. To run Muse Glimmer, download the latest release of Ollama and use the specified commands.
