Developer Tools

Ollama v0.32.7 adds Meta's 30B Muse Glimmer for local agents

Meta's first Superintelligence Labs model runs locally via Ollama on Apple Silicon.

Deep Dive

Ollama released v0.32.7, bringing Muse Glimmer — Meta's newest open model and the first from Meta Superintelligence Labs — to local deployment. Muse Glimmer is a 30B multimodal model designed for agent workloads that run entirely on-device, a departure from cloud-dependent AI agents. The initial release leverages Ollama's MLX engine for state-of-the-art performance on Apple Silicon, with additional optimizations promised for NVIDIA, AMD, and other platforms in the coming days.

Users can immediately run Muse Glimmer with simple commands, such as `ollama run muse-glimmer:30b-mlx`. It integrates with popular coding agents including Claude Code, Codex, and Pi, as well as long-running personal assistant frameworks like OpenClaw and Hermes. Version 0.32.7 also adds support for DFlash and image input, expanding the multimodal capabilities. This release marks a significant step toward practical, private, local AI agents, giving developers a powerful open-source option without sending data to the cloud.

Key Points
  • Muse Glimmer is Meta's first model from Superintelligence Labs — a 30B multimodal model for local agent workloads
  • Initial support via Ollama's MLX engine on Apple Silicon, with NVIDIA/AMD optimizations coming soon
  • Works with Claude Code, Codex, Pi coding agents and OpenClaw, Hermes assistants; includes DFlash and image input

Why It Matters

Local 30B agent models reduce cloud dependency and latency, enabling private, real-time AI assistants on consumer hardware.

📬 Get the top 10 AI stories daily