CRITICAL FIX (2026-03-19): Fixed eos_token_id — previous versions caused infinite thinking loops. You MUST re-download this model if you downloaded before today.

Update (2026-03-18): Models updated to v2.1 with VLM support and fixed configs. Re-download if you got this before today.

MLX Studio — the only app that natively supports JANG models

Early Adoption: LM Studio, Ollama, oMLX, Inferencer do not support JANG yet. Use MLX Studio or pip install "jang[mlx]".

JANG

⚡ All JANG models are meant to be run in vMLX

Qwen3.5-122B-A10B — JANG_3L (3-bit, 8-bit attention) — VLM

JANG — Jang Adaptive N-bit Grading | Mixed-Precision Quantization for Apple Silicon

GitHub

X/Twitter

JANG is fully open-source. Quantization engine, research, and full commit history: github.com/jjang-ai/jangq. Created by Jinho Jang.

Results (200-question MMLU)

Model	MMLU	Size	Speed
JANG_4K	86%	69 GB	~50 tok/s
JANG_3L	81.5%	49 GB	49.6 tok/s
JANG_2S	79%	35 GB	54 tok/s
MLX 4-bit	85%	64 GB	—
MLX 2-bit	56.5%	36 GB	—

JANG_3L: 81.5% at 49 GB — fits on 64 GB Macs. 8-bit attention preserves quality at 3-bit compression.

Per-Subject Scores

Subject	JANG_3L
Abstract Algebra	12/20
Anatomy	18/20
Astronomy	19/20
College CS	15/20
College Physics	15/20
HS Biology	19/20
HS Chemistry	18/20
HS Mathematics	11/20
Logical Fallacies	18/20
World Religions	18/20
Total	163/200 = 81.5%

Specs

Metric	Value
Source	Qwen3.5-122B-A10B
Architecture	MoE (256 experts, 8 active) + GatedDeltaNet SSM
Profile	JANG_3L (CRITICAL=8, IMPORTANT=4, COMPRESS=3)
Average bits	~3.08
GPU Memory	48.5 GB
Speed	49.6 tok/s
VLM	Yes (vision encoder preserved)
Format	v2 (MLX-native, instant load)

Install

pip install "jang[mlx]"

Quick Start

python

from jang_tools.loader import load_jang_model
from mlx_lm.sample_utils import make_sampler
from mlx_lm.generate import generate_step
import mlx.core as mx

model, tokenizer = load_jang_model("JANGQ-AI/Qwen3.5-122B-A10B-JANG_3L")
sampler = make_sampler(temp=0.7)
tokens = tokenizer.encode("What is photosynthesis?")
for tok, _ in generate_step(prompt=mx.array(tokens), model=model, max_tokens=200, sampler=sampler):
    t = tok.item() if hasattr(tok, 'item') else int(tok)
    print(tokenizer.decode([t]), end="", flush=True)
    if t == tokenizer.eos_token_id: break

VLM Inference

python

from jang_tools.loader import load_jang_vlm_model
from mlx_vlm import generate

model, processor = load_jang_vlm_model("JANGQ-AI/Qwen3.5-122B-A10B-JANG_3L")
prompt = processor.tokenizer.apply_chat_template(
    [{"role": "user", "content": [
        {"type": "image", "image": "photo.jpg"},
        {"type": "text", "text": "Describe this image."}
    ]}], add_generation_prompt=True, tokenize=False, enable_thinking=False)
result = generate(model, processor, prompt, ["photo.jpg"], max_tokens=200)
print(result.text)

한국어

Qwen3.5-122B — JANG_3L

JANG은 Apple Silicon을 위한 혼합정밀도 양자화 포맷입니다.

모델	MMLU	크기	속도
JANG_3L	81.5%	49 GB	49.6 tok/s

pip install "jang[mlx]"

GitHub · HuggingFace · MLX Studio

장진호 제작 · Created by Jinho Jang — jangq.ai · @dealignai

JANGQ-AI/Qwen3.5-122B-A10B-JANG_3L