Skip to content
Computer vision

Kimi K3.1 Pro

2.8T MoE с контекстом 1M

1 min readupdated

Moonshot AI выпустила Kimi K3.1 Pro — open-weight нативно мультимодальную agentic-модель и самый мощный представитель 3T-класса у компании на данный момент. Это первая модель серии K3.1: она post-trained из Kimi K3 с крупномасштабным joint reinforcement learning по сотням downstream-доменов.

Ключевые особенности

  • Архитектура MoE: 2.8T параметров всего, 104B активируются на токен.
  • Понимает текст и изображения; vision encoder — MoonViT-V2 на 401M параметров.
  • Контекстное окно 1 048 576 токенов.
  • 93 слоя, 896 экспертов, 16 выбранных экспертов на токен плюс 2 shared.
  • Квантование: веса MXFP4, активации MXFP8, с quantization-aware training.
  • Заметный рост в coding, 3D creation, financial research, factual accuracy и agentic-задачах.

Пример

Сценарий из карточки: подать модели скриншот CAD-интерфейса или 3D-сцены и попросить агента разобрать ошибку, предложить пошаговый фикс и сверить его с длинной документацией в контексте до 1M токенов.

ПодробнееTerminal-Bench 4.0 — 65.7, SWE Marathon v1.1 — 51.7, BenchCAD w/ tools — 92.7.
2.8T total / 104B activated; контекст 1 048 576 токенов.
Веса MXFP4, активации MXFP8 (QAT).
Требований к VRAM в карточке нет.
Discuss Telegram 570 797

Will it run?

Won't run

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

  • U8weights 2643 GB · ready files2907 GB
  • FP8 / INT8recommendedweights 2589 GB2849 GB
  • 4-bitweights 1456 GB1603 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.

What to get for this model

Massed ComputeOpen Massed Compute

No single instance holds the whole model: it needs RAM offload or a cluster.

Your own GPU: Won't fit any home GPU — renting is cheaper.

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

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.

    # moonshotai/Kimi-K3.1-Pro
    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 moonshotai/Kimi-K3.1-Pro --local-dir ~/models/Kimi-K3.1-Pro
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Kimi-K3.1-Pro
  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

New posts about versions of Kimi-K3.1-Pro (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

moonshotai/Kimi-K3.1-Pro
Author
moonshotai
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

797570

Ask AI about this model

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

Comments