mavis-ai/Gemma4-31B-MLX

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 11, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

mavis-ai/Gemma4-31B-MLX is an unmodified redistribution of Google's Gemma 4 31B model, a 31 billion parameter language model. It provides the exact base weights for use in MLX or local AI workflows, maintaining the original model's architecture and capabilities. This model is specifically hosted by mavis-ai to serve as the local reasoning engine for their R.E.V.I.S. Cognitive OS, which enables distributed multi-agentic AI on Mac devices. It is suitable for heavy AI workloads like recursive web research and dynamic RAG generation.

Loading preview...

What is mavis-ai/Gemma4-31B-MLX?

This repository hosts the exact, unmodified base weights of the official Google Gemma 4 31B model. It is a 31 billion parameter language model, provided as a dedicated download source primarily for the mavis-ai application ecosystem, R.E.V.I.S. However, these weights are freely available for general use in MLX or other local AI workflows, being 100% identical to Google's official release.

Key Characteristics & Purpose

  • Unmodified Weights: The model weights are precisely as released by Google, ensuring consistency with the original Gemma 4 31B model's architecture and performance.
  • Local AI Integration: While generally usable, mavis-ai specifically hosts this model to power their R.E.V.I.S. (Local Cognitive OS). R.E.V.I.S. is designed for multi-agentic AI, transforming Mac devices into a distributed agentic swarm for heavy AI workloads.
  • Use Cases within R.E.V.I.S.: When integrated with R.E.V.I.S., this model facilitates advanced tasks such as recursive web research, dynamic RAG (Retrieval Augmented Generation), and multi-step logic processing locally.
  • License: The model is redistributed under the Apache License 2.0, consistent with Google's original licensing for Gemma 4 31B.

Why use this model?

Developers can use this model if they require the official Google Gemma 4 31B weights for local MLX-based development or other local AI applications. It is particularly relevant for those interested in exploring or utilizing mavis-ai's R.E.V.I.S. Cognitive OS for advanced local, distributed AI operations on Mac devices.

Finetunes
5 models