Skip to main content

🦙 Run OpenThai on Your Own Machine with Ollama

Every OpenThai model is free to download and run on your own computer through Ollama — no signup, no API key, no internet needed after the download, and not a single byte of your data leaves your machine.

All OpenThai models on Ollamaollama.com/openthai

1. Install Ollama

Download the installer for macOS, Windows or Linux and install it as usual.

Download Ollamaollama.com/download

2. Pick a model and run it with one command

Open a terminal (PowerShell on Windows) and type the command for the model you want — the first run downloads it automatically.

ModelBest forMin RAM/VRAMCommand
OpenThai 1.5 7BThai chat on an ordinary laptop8 GBollama run openthai/openthai-1.5-7b
OpenThai R1 32BReasoning, maths, code24 GBollama run openthai/openthai-r1-32b
OpenThai 1.6 72BHighest-quality general use48 GBollama run openthai/openthai-1.6-72b
OpenThai 2.0 Legal 30BThai law with section citations24 GBollama run hf.co/iapp/openthai2.0-legal-thaillm-nemotron-3-nano-30b-a3b-GGUF:Q4_K_M

Example:

ollama run openthai/openthai-r1-32b
>>> A shop discounts 20%, then another 15% off the reduced price. Original price 2,500 baht — what do I pay?

The R1 model shows its step-by-step reasoning before answering; the 2.0 Legal model cites the law name and section (มาตรา) alongside its answers.

3. Call it from your own code (API)

Ollama exposes an OpenAI-compatible API at http://localhost:11434 — existing OpenAI SDK code works as-is, just change the base_url.

from openai import OpenAI

client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")

r = client.chat.completions.create(
model="openthai/openthai-r1-32b",
messages=[{"role": "user", "content": "Summarise the advantages of running models locally"}],
)
print(r.choices[0].message.content)

Want the raw model files?

Official GGUF files at every quantization level (Q4_K_M / Q5_K_M / Q8_0) are on Hugging Face, for use with llama.cpp, LM Studio and other tools.

All official GGUF fileshuggingface.co/openthaigpt

Notes

  • The default is Q4_K_M — the best size/quality balance. If you have the memory to spare, the Q8_0 files on Hugging Face give higher quality.
  • ⚠️ Output from the 2.0 Legal model is decision support, not legal advice — verify every cited section against the current statute text.
  • Compare model scores on the Thai LLM Leaderboard
  • Stuck? Ask in Discord