Gargantua Pro Gen 1: our largest model, Gemma 4 12B fine-tuned for openTARS tools, with vision. For computers with room for 8 GB or more.

vision tools thinking

ollama run ProjectOpenTARS/Gargantua-Pro:pro

curl http://localhost:11434/api/chat \
  -d '{
    "model": "ProjectOpenTARS/Gargantua-Pro:pro",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='ProjectOpenTARS/Gargantua-Pro:pro',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'ProjectOpenTARS/Gargantua-Pro:pro',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated yesterday

yesterday

0a5e3ea83c26 · 7.6GB

model

archgemma4

parameters11.9B

quantizationQ4_K_M

7.4GB

projector

archclip

parameters52.4M

quantizationF32

210MB

params

{ "temperature": 0.4, "top_k": 64, "top_p": 0.95 }

44B

Readme

Gargantua Pro Gen 1

Gargantua Pro Gen 1 is the largest model of openTARS: Gemma 4 12B fine-tuned for the openTARS tools, with vision through a projector file included in this package. It needs a computer with plenty of memory, and it is meant for tasks with several steps.

Usage

ollama run ProjectOpenTARS/Gargantua-Pro:pro

The download is about 7.6 GB, model and projector together.

What we have checked

It loads, answers, and calls tools correctly on simple requests. It reads text in an image. It runs on the GPU, and on the CPU more slowly.

Limits

There is no measured score yet. Multi-step tasks are the goal, but they have not been tested at scale. Audio is not included in this build.

Hardware

Works with 16 GB of memory or more, or a graphics card with 8 GB or more. With less memory it will be slow or may not load.

License

The base model is under the Gemma 4 terms (Apache-2.0). The openTARS fine-tune and app are MIT.