OpenTARS/Gargantua-2-Lite

Gargantua 2 Lite (0.8B)

Gargantua 2 Lite is the small model of the openTARS assistant. It is built to run on the CPU, on machines without a graphics card. It opens programs, types, searches the web, runs commands, and can look at a screenshot.

• Base: Qwen3.5 0.8B (Apache-2.0), fine-tuned by the openTARS project on its own tool-use, text-action and screen data, including code examples.
• Files: gargantua-1.2-lite-qwen3.5-0.8b-Q4_0.gguf (0.5 GB) and the vision projector mmproj-gargantua-qwen3.5-0.8b-Q8_0.gguf (0.12 GB).
• Runs with: llama.cpp (llama-server), and with Ollama (ProjectOpenTARS/Gargantua-2:lite).

Measured results (openTARS test suites, local machine)

These are our own tests, run with the openTARS app. They are not public benchmarks.

Test	Result
80-task Lab suite	68/80
Text actions	64/76
Tool-call choice (21 cases)	18/21 (it sometimes picks web search instead of the calculator)
100-task suite with side-effect check (--cem, 1 Nov 2026)	92/100

Known limits: it invents facts more often than the 2B model, and it is weaker at reading screenshots. Use the 2B model when the machine has a graphics card.

Use

With llama.cpp:

llama-server -m gargantua-1.2-lite-qwen3.5-0.8b-Q4_0.gguf --mmproj mmproj-gargantua-qwen3.5-0.8b-Q8_0.gguf -c 8192

With Ollama:

ollama run ProjectOpenTARS/Gargantua-2:lite

Recommended sampling: temperature 0.4, top_p 0.8, top_k 20, repeat_penalty 1.05.

Checksums

See SHA256SUMS. The GGUF file's SHA-256 matches the layer already published on Ollama (a6f164ca…).