mlx-optiq
← Models

Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

mlx-community/Qwen3.5-0.8B-OptiQ-4bit, quantized with OptiQ by ulldma for Apple Silicon.

community
On disk0.7 GB
Memoryabout 2 GB
Parent score38.42 Qwen3.5-0.8B-OptiQ-4bit

This model reuses the per-layer bit allocation of Qwen3.5-0.8B-OptiQ-4bit, whose measured Capability Score is 38.42.

The parent's Capability Score breakdown

This model has no Capability Score of its own. The chart shows the six metrics behind its parent Qwen3.5-0.8B-OptiQ-4bit's score of 38.42, taken from the parent's model card and compared with uniform 4-bit.

  • Qwen3.5-0.8B-OptiQ-4bit
  • Uniform 4-bit
The parent's Capability Score breakdown: Qwen3.5-0.8B-OptiQ-4bitMMLU: Qwen3.5-0.8B-OptiQ-4bit 51.1, Uniform 4-bit 48.5; GSM8K: Qwen3.5-0.8B-OptiQ-4bit 37.3, Uniform 4-bit 31.8; IFEval: Qwen3.5-0.8B-OptiQ-4bit 55.6, Uniform 4-bit 49.5; BFCL-V3: Qwen3.5-0.8B-OptiQ-4bit 55.5, Uniform 4-bit 35.0; HumanEval: Qwen3.5-0.8B-OptiQ-4bit 25.0, Uniform 4-bit 20.1; HashHop: Qwen3.5-0.8B-OptiQ-4bit 6.0, Uniform 4-bit 13.0MMLU51.148.5GSM8K37.331.8IFEval55.649.5BFCL-V355.535.0HumanEval25.020.1HashHop6.013.0

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 ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

OptiQ Code

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

optiq serve --model ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql
optiq code

OptiQ Lab

optiq lab --model ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

Claude Code

optiq run claude -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

Codex

optiq run codex -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

OpenCode

optiq run opencode -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

Mistral Vibe

optiq run vibe -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

Hermes Agent

optiq run hermes -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

OpenClaw

optiq run openclaw -m ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql

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("ulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyQwen · Qwen docs
ArchitectureQwen3.5
Base modelmlx-community/Qwen3.5-0.8B-OptiQ-4bit
Size on disk0.7 GB
Memory neededabout 2 GB
Imagesno, text only
Family referenceThis model reuses the per-layer bit allocation of Qwen3.5-0.8B-OptiQ-4bit, whose measured Capability Score is 38.42.
Downloads54 in the last month
Updated2026-08-23
Hugging Faceulldma/Qwen3.5-0.8B-OptiQ-4bit-text-to-sql