mlx-optiq
← Models

Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit, quantized by OptiQ for Apple Silicon; mixture-of-experts, takes images.

visionMoEexpert-prunedMTP head
On disk14.9 GB
Memoryabout 17 GB
Capability Score76.57
Imagesyes

Capability Score breakdown

The six metrics behind this model's Capability Score of 76.57, as reported in the model card. Values are compared with unpruned parent.

  • This model
  • Unpruned parent
Capability Score breakdown: Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19BMMLU: This model 62.3, Unpruned parent 83.7; GSM8K: This model 90.5, Unpruned parent 87.9; IFEval: This model 76.9, Unpruned parent 72.6; BFCL-V3: This model 91.5, Unpruned parent 92.5; HumanEval: This model 90.2, Unpruned parent 91.5; HashHop: This model 48.0, Unpruned parent 52.0MMLU62.383.7GSM8K90.587.9IFEval76.972.6BFCL-V391.592.5HumanEval90.291.5HashHop48.052.0

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/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

OptiQ Code

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

optiq serve --model mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B
optiq code

OptiQ Lab

optiq lab --model mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

Claude Code

optiq run claude -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

Codex

optiq run codex -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

OpenCode

optiq run opencode -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

Mistral Vibe

optiq run vibe -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

Hermes Agent

optiq run hermes -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

OpenClaw

optiq run openclaw -m mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B

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/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5 MoE
Base modelmlx-community/Qwen3.6-35B-A3B-OptiQ-4bit
Size on disk14.9 GB
Memory neededabout 17 GB
Imagesyes, through optiq serve and the Lab
MTP headincluded
Capability Score76.57 (from the model card)
Downloads2,776 in the last month
Updated2026-08-28
Hugging Facemlx-community/Qwen3.6-35B-A3B-OptiQ-4bit-REAP-19B