Qwen3.8-27B

MLX oQ4e 4-bit, native MTP, vision · omlx · 256K context · Apple M5 Ultra · 256 GB

Alibaba's 27B dense model from August 2026, and the strongest model that fits one consumer card. It thinks before it answers, calls tools reliably and reads images, which makes it the default for coding agents and long projects.

29 tok/sdecode

316 tok/sprefill

256Kcontext window

Earlier checkSep 25, 2026

Run it

Follow the steps below to download and run this recipe. Omarchy Local AI lists the recipes it can launch on your hardware. The server then answers on http://localhost:8000/v1.

1 Install the engine

python3 -m pip install 'omlx @ https://github.com/jundot/omlx/releases/download/v0.7.0.dev4/omlx-0.7.0.dev4-cp313-cp313-macosx_15_0_universal2.whl'

2 Download the weights

hf download Jundot/Qwen3.8-27B-oQ4e-mtp \
  --revision 04dc5509edd8670fc78cc8c2f74bf9b77b1f2acc \
  --local-dir models/Qwen3.8-27B-oQ4e-mtp

3 Write the launcher

mkdir -p omlx-home
cat > omlx-home/model_settings.json <<'EOF'
{
  "version": 1,
  "models": {
    "Qwen3.8-27B-oQ4e-mtp": {
      "mtp_enabled": true
    }
  }
}
EOF

4 Start the server

HF_HUB_OFFLINE=1 \
  omlx \
  serve \
  --model-dir models \
  --base-path omlx-home \
  --host 127.0.0.1 \
  --port 8000

What it passed

✓ loadLoads and serves within an hour

✓ chatAnswers a plain question and stops on its own

– reasoningThinks separately and gets 17 × 23 right

– toolsCalls a tool with the right arguments and uses the result

✓ contextRecalls a code buried in a prompt that fills 85% of the window

– speedDecodes at 15 tok/s or more

Passed the older check (loads and chats); a full six-check run is pending.

Details

Weights: Jundot/Qwen3.8-27B-oQ4e-mtp @ 04dc5509ed ›
Image: none: a program on the host
Launch: omlx-qwen3.8-27b-oq4e-mtp-256k ›
Recipe file: registry/recipes/apple/apple-m5-ultra-256gb/qwen3.8-27b.omlx.256k.json ›
Model released: Aug 5, 2026
Tested: earlier acceptance