Skip to content
Language models

Qwen3.8-Flash-Next-Brisk-Coder

an optimized version for programming and agent tasks

2 min read

The model is a specialized version of the base Brisk architecture, created by pruning half the experts in each layer using a mask from ISTA-DASLab. The author kept 256 out of 512 experts, optimizing the model for coding and agent workflows while maintaining high generation speed and the full Multi-Token Prediction block. By reducing active experts, the model requires less memory to run while retaining accuracy on complex programming tasks. It is well suited for developers using local environments on Apple Silicon to write and debug code.

Key features

  • Reduced memory usage: routed experts take up half as much memory (about 44 GB with an n-gram table on SSD instead of 86 GB for the original).
  • Coding specialization: expert masks applied for coding, agent scenarios, and visual data processing.
  • Retained MTP block: the full multi-token prediction block remains unchanged to maintain high generation speed.
  • High performance: the model demonstrates excellent results on code generation benchmarks (LiveCodeBench).

What you can do

  • Write and debug code: use the model in local environments (such as oMLX) to generate code in various programming languages with thinking mode enabled.
  • Test agent tasks: apply the model to automate development tasks requiring deep context analysis.
  • Integrate into local workflows: run the model on Mac with Apple Silicon support for efficient local development without sending data to the cloud.

Example

hf download blazkepa/Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP \
  --local-dir ~/.omlx/models/blazkepa/Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP

omlx restart
Read moreDetails
Benchmarks: LiveCodeBench pass@1 — 77% (compared to 63% for the base model and 68% for the version without LoRA).
Hardware requirements: about 79 GB of RAM on a 128 GB Mac or about 44 GB when offloading the n-gram table to SSD.
Limitations: quality drops outside of coding (MMLU drops from 89 to 77%), Polish chat is broken.
Library: mlx, overdrive via oMLX 0.8.0 on macOS.
Discuss Telegram 0 17

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • U32weights 223 GB · ready files224 GB
  • FP8 / INT8recommendedweights 111 GB113 GB
  • 4-bitweights 63 GB64 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.

Language models: with a 8192-token context.

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

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent 4× RTX A6000

    $1.71 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # blazkepa/Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP — 4× RTX A6000, FP8 / INT8
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve blazkepa/Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP --max-model-len 8192 --quantization fp8 --tensor-parallel-size 4
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

  1. Qwen3.8-Flash-Nextfamily root
  2. Qwen3.8-Flash-Next-oQ4e-WLSRefit-MTP
  3. Qwen3.8-Flash-Next-norev-oQ4e-MTP
  4. Qwen3.8-Flash-Next-Brisk-oQ4e-MTP
  5. Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP

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

Quantised versions 12All on Hugging Face
Fine-tunes 8All on Hugging Face

About the model

blazkepa/Qwen3.8-Flash-Next-Brisk-Coder-oQ4e-MTP
Author
blazkepa
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

170

Ask AI about this model

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

Comments