MoziAI-35B-V3.8
A locally deployable, open-source multimodal mixture-of-experts language model focused on financial applications, compressed to under 16 GB through a custom quantization method while running on consumer GPUs.
MoziAI-35B-V3.8 is built on an open-source mixture-of-experts base model, using a Qwen3.5/Qwen3.6-35B-A3B architecture, fine-tuned by its developers with financial domain data, a structured multi-step reasoning framework, and an agent reflection-and-iteration mechanism referred to as LOOP.
The 35B-parameter mixture-of-experts model is compressed to about 15.9 GB through the developers' own quantization method, around 30% smaller than a standard Q4_K_M quantization, letting it run on consumer GPUs with 20 GB of VRAM or more for local, offline inference. It supports vision input and tool calling.
Released in September 2026, the model is licensed under Apache-2.0 for its own components, building on Apache-2.0-licensed base components and an MIT-licensed upstream increment, with license details recorded in the repository's NOTICE and LICENSE files.