Ollama Adds Support for Meta's 30B "Muse Glimmer" Model, First Release from Meta Superintelligence Labs
Ollama v0.32.7 adds support for Muse Glimmer, a 30B-parameter multimodal model that Meta describes as the first release from its new Meta Superintelligence Labs division. Initial support runs through Ollama's MLX engine on Apple Silicon, with NVIDIA and AMD support promised in the coming days.
Meta Superintelligence Labs Ships Its First Model
Ollama version 0.32.7 adds support for Muse Glimmer, a 30B-parameter multimodal model that, according to Ollama's release notes, is "Meta's newest open model and the first released by Meta Superintelligence Labs." The model is designed specifically for agent workloads that run on local hardware rather than through a cloud API.
As of this release, Muse Glimmer is available exclusively through Ollama's MLX engine on Apple Silicon. Ollama states that support for NVIDIA, AMD, and other platforms is coming "in the coming days," meaning non-Mac users cannot yet run the model through Ollama.
What's New in the Runtime
Ollama 0.32.7 adds two specific capabilities to its MLX engine required to run this model:
- DFlash support
- Image input support
No context window size, benchmark scores, or parameter-level architecture details (e.g., dense vs. mixture-of-experts) were disclosed in the release notes. Pricing does not apply in the traditional API sense since this is a locally-run open-weight model — there are no published per-token input/output costs.
How to Run It
Users can pull and run the model directly:
ollama run muse-glimmer:30b-mlx
Ollama is positioning Muse Glimmer as a backend for existing agent tooling rather than a standalone chatbot. The release notes list explicit integration commands for coding agents and personal-assistant frameworks:
ollama launch claude --model muse-glimmer:30b-mlx
ollama launch pi --model muse-glimmer:30b-mlx
ollama launch openclaw --model muse-glimmer:30b-mlx
ollama launch hermes --model muse-glimmer:30b-mlx
These commands wire the model into Claude Code, Codex-style coding agents, the lightweight Pi agent, and long-running personal-assistant frameworks OpenClaw and Hermes.
What This Means
This release marks a notable data point: Meta Superintelligence Labs, the internal group Meta formed to consolidate its frontier AI efforts, has apparently pushed its first public artifact out as an open, locally-runnable model rather than a hosted product. The emphasis on agent workloads, image input, and integration with coding-agent frameworks like Claude Code and Codex suggests Meta is targeting developers building autonomous coding and assistant tools rather than general chat use cases.
The rollout is incomplete — Apple Silicon only, no NVIDIA or AMD support yet, and no independent benchmark data to verify capability claims. Until Meta or Meta Superintelligence Labs publishes technical documentation, benchmark scores, and a model card, claims about the model's quality and multimodal performance remain unverified. Developers running local agent stacks on Macs can test the model today; everyone else will need to wait for broader hardware support.
Related Articles
inclusionAI releases Duo 1.0 causal-base, a full-duplex multimodal world model initialized from MiniMax H3
inclusionAI has published the Duo 1.0 causal-base weights on Hugging Face. Duo is described as a full-duplex multimodal world model for real-time audio-visual interaction, initialized from MiniMax H3 and trained further by the Duo team.
Translated releases Lara Mini, a 7.2B-parameter MoE research model with 600M active parameters
Translated announced Lara Mini on November 12, 2026, a research model trained from scratch on 6 trillion tokens across English and 44 other languages, including all 24 official EU languages. The mixture-of-experts system has 7.2 billion parameters and activates 600 million per token.
Endless Frontier open-sources Fx-Bio-1031, a 125B MoE bioinformatics model with 512K context
Endless Frontier (无界前沿) has open-sourced Fx-Bio-1031, a bioinformatics model with 125B total parameters and 6B active per token, with a 512K context window. The company claims 70.59% Pass@5 on the Difficult subset of Anthropic's BioMysteryBench, against 47.06% for GPT-6-Astra and 52.94% for Claude Opus-5.
Huawei Cloud to unveil CloudRobo-WM1.2, its first Ascend-native world model, on November 16
Huawei Cloud will launch CloudRobo-WM1.2, an embodied world model for robots, at an online event on November 16 (14:00 to 15:30 Beijing time). Its poster bills the model as the first Ascend-native world model from the Industry AI Dream Factory. Parameter counts, benchmarks, availability and pricing have not been disclosed.
Comments
Loading...