mlx-optiq
← Models

Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

mistralai/Devstral-Small-2-24B-Instruct-2512, quantized by OptiQ for Apple Silicon; takes images.

vision
On disk17.4 GB
Memoryabout 21 GB
Capability Score78.60
Imagesyes

Capability Score breakdown

The six metrics behind this model's Capability Score of 78.60, as reported in the model card. Values are compared with uniform 4-bit.

  • This model
  • Uniform 4-bit
Capability Score breakdown: Devstral-Small-2-24B-Instruct-2512-OptiQ-4bitMMLU: This model 80.8, Uniform 4-bit 79.4; GSM8K: This model 87.2, Uniform 4-bit 88.1; IFEval: This model 68.6, Uniform 4-bit 68.0; BFCL-V3: This model 89.5, Uniform 4-bit 88.5; HumanEval: This model 83.5, Uniform 4-bit 82.9; HashHop: This model 62.0, Uniform 4-bit 53.0MMLU80.879.4GSM8K87.288.1IFEval68.668.0BFCL-V389.588.5HumanEval83.582.9HashHop62.053.0

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/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

OptiQ Code

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

optiq serve --model mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit
optiq code

OptiQ Lab

optiq lab --model mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

Claude Code

optiq run claude -m mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

Codex

optiq run codex -m mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

OpenCode

optiq run opencode -m mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

Mistral Vibe

optiq run vibe -m mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

Hermes Agent

optiq run hermes -m mlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit

OpenClaw

optiq run openclaw -m mlx-community/Devstral-Small-2-24B-Instruct-2512-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/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit")
print(generate(model, tokenizer, prompt="Hello", max_tokens=256))

Details

FamilyMistral · Devstral · Mistral · Devstral docs
ArchitectureMistral 3
Base modelmistralai/Devstral-Small-2-24B-Instruct-2512
Size on disk17.4 GB
Memory neededabout 21 GB
Imagesyes, through optiq serve and the Lab
Capability Score78.60 (from the model card)
Downloads1,158 in the last month
Updated2026-09-24
Hugging Facemlx-community/Devstral-Small-2-24B-Instruct-2512-OptiQ-4bit