Instructions to use mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp")
config = load_config("mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MiniMax-M3.1-Flash-oQ4e-mtp

This model was quantized using oQ (oMLX v0.7.3.dev1) mixed-precision quantization.

A mixed-precision MLX checkpoint of MiniMaxAI/MiniMax-M3.1-Flash, a native multimodal mixture-of-experts model for coding, agentic workflows, computer use and cybersecurity with about 428B total and 23B activated parameters. It takes text, image and video inputs with a context window of up to 1M tokens, and uses MiniMax Sparse Attention throughout the model. It includes the language model, the vision encoder and the bundled DSpark draft in dspark/, which oMLX Lightning MTP uses for speculative decoding.

The effective average is 4.617 bits/weight for the language model and vision encoder, or 4.638 bits/weight including the DSpark draft.

Quantization and Bit Distribution

oQ4e keeps the routed experts in their source NVFP4 format. It keeps MXFP8 for the shared experts, the sparse attention indexer and more than half of the attention projections, and stores the remaining projections, the embeddings and the vocabulary head as affine 4- to 8-bit. The router and the DSpark draft keep their source precision.

Storage format	Logical weights	Share of total	Tensor storage	Effective bits/weight
Affine 4-bit, group 64	0.189B	0.044%	0.099 GiB	4.50
NVFP4, group 16	413.122B	96.738%	216.422 GiB	4.50
Affine 5-bit, group 64	2.378B	0.557%	1.523 GiB	5.50
Affine 6-bit, group 64	1.085B	0.254%	0.821 GiB	6.50
Affine 8-bit, group 64	2.458B	0.576%	2.433 GiB	8.50
MXFP8, group 32	7.097B	1.662%	6.816 GiB	8.25
BF16	0.678B	0.159%	1.262 GiB	16.00
FP32	0.045B	0.011%	0.167 GiB	32.00
Total	427.052B	100%	229.542 GiB	4.617

Storage Breakdown

Component	Logical weights	Storage
Language model	426.375B	245.116 GB / 228.282 GiB
Vision	0.677B	1.354 GB / 1.261 GiB
DSpark draft (MTP)	2.011B	2.302 GB / 2.144 GiB
Total	429.063B	248.771 GB / 231.686 GiB

License

MiniMax Community License, following the source model. See LICENSE. Commercial use has additional conditions in the license, including displaying "Built with MiniMax M3.1-Flash" and obtaining prior written authorization from MiniMax.

Downloads last month: 582

Safetensors

Model size

427B params

Tensor type

U32

BF16

F32

MLX

Hardware compatibility

4-bit

Model tree for mlx-community/MiniMax-M3.1-Flash-oQ4e-mtp

Base model

MiniMaxAI/MiniMax-M3.1-Flash

Quantized

(32)

this model