mlx-optiq
← Models

Qwen3.5-2B-OptiQ-4bit-text-mlx

mlx-community/Qwen3.5-2B-OptiQ-4bit, quantized with OptiQ by anerjy for Apple Silicon.

community
On disk1.5 GB
Memoryabout 3 GB
Parent score50.41 Qwen3.5-2B-OptiQ-4bit

This model reuses the per-layer bit allocation of Qwen3.5-2B-OptiQ-4bit, whose measured Capability Score is 50.41.

The parent's Capability Score breakdown

This model has no Capability Score of its own. The chart shows the six metrics behind its parent Qwen3.5-2B-OptiQ-4bit's score of 50.41, taken from the parent's model card and compared with uniform 4-bit.

  • Qwen3.5-2B-OptiQ-4bit
  • Uniform 4-bit
The parent's Capability Score breakdown: Qwen3.5-2B-OptiQ-4bitMMLU: Qwen3.5-2B-OptiQ-4bit 58.9, Uniform 4-bit 58.6; GSM8K: Qwen3.5-2B-OptiQ-4bit 55.6, Uniform 4-bit 56.4; IFEval: Qwen3.5-2B-OptiQ-4bit 59.7, Uniform 4-bit 58.6; BFCL-V3: Qwen3.5-2B-OptiQ-4bit 77.0, Uniform 4-bit 74.0; HumanEval: Qwen3.5-2B-OptiQ-4bit 51.2, Uniform 4-bit 39.6; HashHop: Qwen3.5-2B-OptiQ-4bit 0.0, Uniform 4-bit 0.0MMLU58.958.6GSM8K55.656.4IFEval59.758.6BFCL-V377.074.0HumanEval51.239.6HashHop0.00.0

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 anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

OptiQ Code

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

optiq serve --model anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx
optiq code

OptiQ Lab

optiq lab --model anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

Claude Code

optiq run claude -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

Codex

optiq run codex -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

OpenCode

optiq run opencode -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

Mistral Vibe

optiq run vibe -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

Hermes Agent

optiq run hermes -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

OpenClaw

optiq run openclaw -m anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx

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("anerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelmlx-community/Qwen3.5-2B-OptiQ-4bit
Size on disk1.5 GB
Memory neededabout 3 GB
Imagesno, text only
Family referenceThis model reuses the per-layer bit allocation of Qwen3.5-2B-OptiQ-4bit, whose measured Capability Score is 50.41.
Downloads16 in the last month
Updated2026-08-01
Hugging Faceanerjy/Qwen3.5-2B-OptiQ-4bit-text-mlx