mlx-optiq
← Models

qwen3_5_0_8b_optiq_4bit

Qwen/Qwen3.5-0.8B, quantized with OptiQ by tbhrc for Apple Silicon.

community
On disk0.7 GB
Memoryabout 2 GB
Capability Score41.60

Capability Score breakdown

The six metrics behind this model's Capability Score of 41.60, as reported in the model card.

Capability Score breakdown: qwen3_5_0_8b_optiq_4bitMMLU: This model 54.5; GSM8K: This model 37.3; IFEval: This model 45.8; BFCL-V3: This model 43.0; HumanEval: This model 27.4MMLU54.5GSM8K37.3IFEval45.8BFCL-V343.0HumanEval27.4

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 tbhrc/qwen3_5_0_8b_optiq_4bit

OptiQ Code

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

optiq serve --model tbhrc/qwen3_5_0_8b_optiq_4bit
optiq code

OptiQ Lab

optiq lab --model tbhrc/qwen3_5_0_8b_optiq_4bit

Claude Code

optiq run claude -m tbhrc/qwen3_5_0_8b_optiq_4bit

Codex

optiq run codex -m tbhrc/qwen3_5_0_8b_optiq_4bit

OpenCode

optiq run opencode -m tbhrc/qwen3_5_0_8b_optiq_4bit

Mistral Vibe

optiq run vibe -m tbhrc/qwen3_5_0_8b_optiq_4bit

Hermes Agent

optiq run hermes -m tbhrc/qwen3_5_0_8b_optiq_4bit

OpenClaw

optiq run openclaw -m tbhrc/qwen3_5_0_8b_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("tbhrc/qwen3_5_0_8b_optiq_4bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelQwen/Qwen3.5-0.8B
Size on disk0.7 GB
Memory neededabout 2 GB
Imagesno, text only
Capability Score41.60 (from the model card)
Downloads14 in the last month
Updated2026-08-22
Hugging Facetbhrc/qwen3_5_0_8b_optiq_4bit