mlx-optiq
← Models

Fara1.5-4B-OptiQ-4bit

microsoft/Fara1.5-4B, quantized by OptiQ for Apple Silicon; takes images.

vision
On disk4.2 GB
Memoryabout 6 GB
Parent score68.76 Qwen3.5-4B-OptiQ-4bit
Imagesyes

This model reuses the per-layer bit allocation of Qwen3.5-4B-OptiQ-4bit, whose measured Capability Score is 68.76.

The parent's Capability Score breakdown

This model has no Capability Score of its own. The chart shows the six metrics behind its parent Qwen3.5-4B-OptiQ-4bit's score of 68.76, taken from the parent's model card and compared with uniform 4-bit.

  • Qwen3.5-4B-OptiQ-4bit
  • Uniform 4-bit
The parent's Capability Score breakdown: Qwen3.5-4B-OptiQ-4bitMMLU: Qwen3.5-4B-OptiQ-4bit 69.9, Uniform 4-bit 68.7; GSM8K: Qwen3.5-4B-OptiQ-4bit 80.5, Uniform 4-bit 78.8; IFEval: Qwen3.5-4B-OptiQ-4bit 69.1, Uniform 4-bit 68.4; BFCL-V3: Qwen3.5-4B-OptiQ-4bit 90.0, Uniform 4-bit 85.0; HumanEval: Qwen3.5-4B-OptiQ-4bit 78.0, Uniform 4-bit 76.2; HashHop: Qwen3.5-4B-OptiQ-4bit 25.0, Uniform 4-bit 24.0MMLU69.968.7GSM8K80.578.8IFEval69.168.4BFCL-V390.085.0HumanEval78.076.2HashHop25.024.0

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/Fara1.5-4B-OptiQ-4bit

OptiQ Code

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

optiq serve --model mlx-community/Fara1.5-4B-OptiQ-4bit
optiq code

OptiQ Lab

optiq lab --model mlx-community/Fara1.5-4B-OptiQ-4bit

Claude Code

optiq run claude -m mlx-community/Fara1.5-4B-OptiQ-4bit

Codex

optiq run codex -m mlx-community/Fara1.5-4B-OptiQ-4bit

OpenCode

optiq run opencode -m mlx-community/Fara1.5-4B-OptiQ-4bit

Mistral Vibe

optiq run vibe -m mlx-community/Fara1.5-4B-OptiQ-4bit

Hermes Agent

optiq run hermes -m mlx-community/Fara1.5-4B-OptiQ-4bit

OpenClaw

optiq run openclaw -m mlx-community/Fara1.5-4B-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/Fara1.5-4B-OptiQ-4bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelmicrosoft/Fara1.5-4B
Size on disk4.2 GB
Memory neededabout 6 GB
Imagesyes, through optiq serve and the Lab
Family referenceThis model reuses the per-layer bit allocation of Qwen3.5-4B-OptiQ-4bit, whose measured Capability Score is 68.76.
Downloads134 in the last month
Updated2026-10-18
Hugging Facemlx-community/Fara1.5-4B-OptiQ-4bit