Skip to content
Computer vision

Qwen3.8-27B-opencode-adapted (MLX)

an optimized model for Apple Silicon

2 min readupdated

Built on the Qwen3.8-27B base model and fine-tuned with the AsyncGRPO algorithm, this build is tailored for the OpenCode coding agent. Converted to the MLX format for Apple Silicon, it retains image input support and includes a built-in Multi-Token Prediction (MTP) layer for faster generation. It excels at software engineering, terminal tasks, and technical documentation, making it a strong local option for developers and engineers automating their coding workflows.

Key features

  • Fine-tuned by harness-hub specifically for software engineering and terminal workflows.
  • Full hardware acceleration on Apple Silicon chips via the MLX library.
  • Retained vision encoder supporting image-text-to-text prompts.
  • Built-in MTP (Multi-Token Prediction) layer for faster text decoding.

What you can do

  • Spin up a local OpenAI-compatible server via mlx-lm to integrate with coding agents.
  • Write a Python script to generate code and troubleshoot technical tasks directly on a Mac.
  • Analyze terminal or UI error screenshots using mlx-vlm.
  • Configure speculative decoding with the built-in MTP module to boost generation speed.

Example

Launch a local server to run the model from the terminal:

pip install -U "mlx-lm>=0.32.2"
mlx_lm.server --model netprojects/Qwen3.8-27B-opencode-adapted-MLX-bf16 --port 8000 \ --chat-template-args '{"enable_thinking": false}'
Read moreDetails
Model size (original bf16): 55.6 GB (4-bit, 6-bit, and 8-bit quantized versions are also available).
Benchmark scores (on the original adapter): SWE-bench Pro (hard 51) jumped from 35.3% to 51.0%, and Terminal-Bench 2.1 went from 46.4% to 52.4%.
License: Apache-2.0.
Discuss Telegram 0 36

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 52 GB · ready files57 GB
  • FP8 / INT8recommendedweights 26 GB29 GB
  • 4-bitweights 15 GB17 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # netprojects/Qwen3.8-27B-opencode-adapted-MLX-bf16
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download netprojects/Qwen3.8-27B-opencode-adapted-MLX-bf16 --local-dir ~/models/Qwen3.8-27B-opencode-adapted-MLX-bf16
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Qwen3.8-27Bfamily root
  2. Qwen3.8-27B-opencode-adapted
  3. Qwen3.8-27B-opencode-adapted-MLX-bf16

New posts about versions of Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

netprojects/Qwen3.8-27B-opencode-adapted-MLX-bf16
Author
netprojects
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

360

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments