Gemma 4 26B is Google's model with a Mixture-of-Experts design: only part of its parameters work on each step, so answers are usually quicker than from a dense model of the same size. It understands images, so you can attach a photo, a screenshot or a diagram and ask about it. The template runs it on a card with 24 GB or more, with a chat and an API.
In short
Gemma 4 26B runs on cards with 24 GB or more. The most affordable suitable card right now is the Tesla V100 32 GB from 23 ₽ per hour, billed by the minute.
Updated 13 October 2026
You need a card with at least 24 GB of video memory: the weights take about 18 GB, and images and the context need room on top. An RTX 4090 with 24 GB or an RTX 3090 will run it. An RTX 5090 with 32 GB leaves room for long chats and large images.
Prices and availability come from the live catalog and update automatically.
Attach a screenshot, photo or diagram to your message and ask “What is wrong here?” or “Turn this table into text.” The model answers from the image. It usually reads the text in pictures too, but small print is best uploaded at full resolution, not compressed by a messenger.
Images use up part of the context, so a few large screenshots in one conversation fill it quickly. Start a new chat for a new set of images.
The template's context is 16,384 tokens, about twenty pages of text. The model sees only what fits into it, so split a long file into parts and ask about each section. For how to size the memory for a context, see the VRAM guide for LLMs.
The API lives at https://<id>-11434.run.gpu.nz/v1 and follows the OpenAI format. To check that the server answers, list the models: the response names the model gemma4:26b, which goes into the model field. The token is sent in the Authorization header:
curl "https://<id>-11434.run.gpu.nz/v1/models" \
-H "Authorization: Bearer <token>"If you want another Ollama model in the same setup, see Open WebUI + Ollama.