Gemma 4 26B A4B

· llama.cpp · 128K context · RTX 5070 · 12 GB

Google's Gemma 4 mixture-of-experts model: 26B total, 4B active, so it runs fast while seeing images and thinking.

35 tok/sdecode

1,460 tok/sprefill

128Kcontext window

Tested on this cardOct 8, 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 Download the weights

hf download unsloth/gemma-4-26B-A4B-it-GGUF gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf mmproj-F16.gguf mtp-gemma-4-26B-A4B-it.gguf \
  --revision c099eb48e663fd284577b04978a94ffccb261841 \
  --local-dir ~/models/gemma-4-26B-A4B-it-GGUF-c099eb48

2 Start the server

docker run --rm \
  --gpus all \
  -p 8000:8080 \
  --shm-size 16g \
  -e CUDA_DEVICE_ORDER=PCI_BUS_ID \
  -v ~/models/gemma-4-26B-A4B-it-GGUF-c099eb48:/models:ro \
  --entrypoint /app/llama-server \
  ghcr.io/ggml-org/llama.cpp:server-cuda12-b11146@sha256:52192a23c7b258044463aebba78826cdff39a7393b713edc313324b343a7b836 \
  --model /models/gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf \
  --mmproj /models/mmproj-F16.gguf \
  --spec-type draft-mtp \
  --model-draft /models/mtp-gemma-4-26B-A4B-it.gguf \
  --spec-draft-n-max 3 \
  --cache-type-k q8_0 \
  --cache-type-v q8_0 \
  --load-mode none \
  --host 0.0.0.0 \
  --port 8080 \
  --ctx-size 131072 \
  --fit-target 2560 \
  --flash-attn on \
  --parallel 1 \
  --jinja \
  --cache-ram 0 \
  --metrics \
  --chat-template-kwargs '{"enable_thinking":true}' \
  --temp 1.0 \
  --top-p 0.95 \
  --top-k 64

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 all six checks on a real RTX 5070 on Oct 8, 2026.

Details

Weights: unsloth/gemma-4-26B-A4B-it-GGUF @ c099eb48e6 ›
Image: ghcr.io/ggml-org/llama.cpp:server-cuda12-b11146@sha256:52192a23c7b258044463aebba78826cdff39a7393b713edc313324b343a7b836
Launch: llama.cpp-gemma-4-26b-a4b-ud-q4-k-xl-mtp-autofit-128k-rtx-5070-12gb ›
Recipe file: registry/recipes/nvidia/rtx-5070-12gb/gemma-4-26b-a4b.llama.cpp.128k.json ›
Model released: Mar 12, 2026
Tested: owner, RTX 5070