mlx-optiq
← Models

Qwythos-9B-v2-OptiQ-4bit

empero-ai/Qwythos-9B-v2, quantized by OptiQ for Apple Silicon; takes images.

visionMTP head
On disk8.2 GB
Memoryabout 11 GB
Parent score69.85 Qwen3.5-9B-OptiQ-4bit
Imagesyes

This model reuses the per-layer bit allocation of Qwen3.5-9B-OptiQ-4bit, whose measured Capability Score is 69.85.

The parent's Capability Score breakdown

This model has no Capability Score of its own. The chart shows the six metrics behind its parent Qwen3.5-9B-OptiQ-4bit's score of 69.85, taken from the parent's model card and compared with uniform 4-bit.

  • Qwen3.5-9B-OptiQ-4bit
  • Uniform 4-bit
The parent's Capability Score breakdown: Qwen3.5-9B-OptiQ-4bitMMLU: Qwen3.5-9B-OptiQ-4bit 69.1, Uniform 4-bit 68.6; GSM8K: Qwen3.5-9B-OptiQ-4bit 81.7, Uniform 4-bit 81.7; IFEval: Qwen3.5-9B-OptiQ-4bit 71.2, Uniform 4-bit 71.0; BFCL-V3: Qwen3.5-9B-OptiQ-4bit 91.0, Uniform 4-bit 92.0; HumanEval: Qwen3.5-9B-OptiQ-4bit 81.1, Uniform 4-bit 78.7; HashHop: Qwen3.5-9B-OptiQ-4bit 25.0, Uniform 4-bit 26.0MMLU69.168.6GSM8K81.781.7IFEval71.271.0BFCL-V391.092.0HumanEval81.178.7HashHop25.026.0

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/Qwythos-9B-v2-OptiQ-4bit

OptiQ Code

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

optiq serve --model mlx-community/Qwythos-9B-v2-OptiQ-4bit
optiq code

OptiQ Lab

optiq lab --model mlx-community/Qwythos-9B-v2-OptiQ-4bit

Claude Code

optiq run claude -m mlx-community/Qwythos-9B-v2-OptiQ-4bit

Codex

optiq run codex -m mlx-community/Qwythos-9B-v2-OptiQ-4bit

OpenCode

optiq run opencode -m mlx-community/Qwythos-9B-v2-OptiQ-4bit

Mistral Vibe

optiq run vibe -m mlx-community/Qwythos-9B-v2-OptiQ-4bit

Hermes Agent

optiq run hermes -m mlx-community/Qwythos-9B-v2-OptiQ-4bit

OpenClaw

optiq run openclaw -m mlx-community/Qwythos-9B-v2-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/Qwythos-9B-v2-OptiQ-4bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelempero-ai/Qwythos-9B-v2
Size on disk8.2 GB
Memory neededabout 11 GB
Imagesyes, through optiq serve and the Lab
MTP headincluded
Family referenceThis model reuses the per-layer bit allocation of Qwen3.5-9B-OptiQ-4bit, whose measured Capability Score is 69.85.
Downloads463 in the last month
Updated2026-10-18
Hugging Facemlx-community/Qwythos-9B-v2-OptiQ-4bit