Instructions to use Duckets/Sage-A1.5-MLX-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Duckets/Sage-A1.5-MLX-4bit 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("Duckets/Sage-A1.5-MLX-4bit")
config = load_config("Duckets/Sage-A1.5-MLX-4bit")

# 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 Duckets/Sage-A1.5-MLX-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Duckets/Sage-A1.5-MLX-4bit"

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": "Duckets/Sage-A1.5-MLX-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use Duckets/Sage-A1.5-MLX-4bit 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 "Duckets/Sage-A1.5-MLX-4bit"

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 Duckets/Sage-A1.5-MLX-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Duckets/Sage-A1.5-MLX-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Duckets/Sage-A1.5-MLX-4bit"

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 "Duckets/Sage-A1.5-MLX-4bit" \
  --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"

Sage-A1.5 — MLX 4-bit (vision complete)

MLX 4-bit conversion of InternScience/Sage-A1.5 by InternScience / Shanghai AI Laboratory. All credit for the model goes to them — this repo is only a format conversion for Apple Silicon.

Sage-A1.5 is a 35B MoE agentic model for long-horizon scientific problem solving, released 2026-10-21. This conversion was built and published the same day.

Conversion details

Converted with mlx-vlm 0.7.10 from the original BF16 weights
Affine quantization, 4 bits / group size 64 (router gates at 8-bit, mlx-vlm default recipe) — 4.649 bits per weight average
Vision tower included (333 vision tensors) — image input works, this is not a text-only strip
The separate MTP draft weights (model-mtp.safetensors in the original repo) are not included: LM Studio cannot load the qwen3_5_mtp draft architecture, so a bundle including it would not run there. This package is the complete main model.
262,144 token native context

Use in LM Studio

Download the folder, place it in your LM Studio models directory, restart LM Studio so it indexes the folder, and load it as a vision (image-text-to-text) model.

Headline benchmarks (from the original model card)

Benchmark	Sage-A1.5	Ornith-1.5 (~35B)
GAIA	95.1	77.6
IFBench	81.6	44.6
BrowseComp	81.7	67.6
SWE-bench Pro	61.2	59.6
FrontierScience-Research	48.3	6.7
HLE w/ tools	55.1	33.4

Full evaluation table: see the original model card.

License

Apache-2.0, same as the original model.

Converted by Duckets — local AI on Apple Silicon. 🦆

Downloads last month: -

Safetensors

Model size

35B params

Tensor type

U32

BF16

MLX

Hardware compatibility

4-bit

Model tree for Duckets/Sage-A1.5-MLX-4bit

Base model

InternScience/Sage-A1.5

Quantized

(19)

this model