mlx-optiq
← Models

Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

DavidAU/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking, quantized by OptiQ for Apple Silicon.

On disk18.7 GB
Memoryabout 23 GB
Parent score86.04 Qwen3.6-27B-OptiQ-4bit

This model reuses the per-layer bit allocation of Qwen3.6-27B-OptiQ-4bit, whose measured Capability Score is 86.04.

The parent's Capability Score breakdown

This model has no Capability Score of its own. The chart shows the six metrics behind its parent Qwen3.6-27B-OptiQ-4bit's score of 86.04, taken from the parent's model card and compared with uniform 4-bit.

  • Qwen3.6-27B-OptiQ-4bit
  • Uniform 4-bit
The parent's Capability Score breakdown: Qwen3.6-27B-OptiQ-4bitMMLU: Qwen3.6-27B-OptiQ-4bit 87.4, Uniform 4-bit 87.6; GSM8K: Qwen3.6-27B-OptiQ-4bit 92.0, Uniform 4-bit 92.1; IFEval: Qwen3.6-27B-OptiQ-4bit 74.1, Uniform 4-bit 71.7; BFCL-V3: Qwen3.6-27B-OptiQ-4bit 92.5, Uniform 4-bit 93.0; HumanEval: Qwen3.6-27B-OptiQ-4bit 90.2, Uniform 4-bit 92.1; HashHop: Qwen3.6-27B-OptiQ-4bit 80.0, Uniform 4-bit 77.0MMLU87.487.6GSM8K92.092.1IFEval74.171.7BFCL-V392.593.0HumanEval90.292.1HashHop80.077.0

Use it

Install

pip install mlx-optiq

Serve it

An OpenAI- and Anthropic-compatible server on this Mac. Every app below can use it.

optiq serve --model mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

OptiQ Code

The terminal coding agent uses whatever model optiq serve is running.

optiq serve --model mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit
optiq code

OptiQ Lab

optiq lab --model mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

Claude Code

optiq run claude -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

Codex

optiq run codex -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

OpenCode

optiq run opencode -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

Mistral Vibe

optiq run vibe -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

Hermes Agent

optiq run hermes -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

OpenClaw

optiq run openclaw -m mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit

optiq run starts optiq serve if it is not running, sets the agent up for this session only, and stops the server when the agent exits.

mlx-lm, text only

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelDavidAU/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking
Size on disk18.7 GB
Memory neededabout 23 GB
Imagesno, text only
Family referenceThis model reuses the per-layer bit allocation of Qwen3.6-27B-OptiQ-4bit, whose measured Capability Score is 86.04.
Downloads562 in the last month
Updated2026-09-16
Hugging Facemlx-community/Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking-OptiQ-4bit