OpenThai 2.0 Legal

What it does​

• Cites the sectionAnswers in Thai with the law name and มาตรา it relied on, as JSON your code can check.
• Knows the statutesWith no law in the prompt, roughly four times the recall of Qwen3.6-35B on the Civil and Commercial Code.
• Writes legal analysisAhead of Qwen3.6-35B on citations, holding, coverage and fluency, graded on Thai Supreme Court cases.
• Runs on your hardwareOne 24 GB GPU with the 4-bit weights, so case files stay inside your network.

Try it​

curl -s https://api.iapp.co.th/v3/llm/openthai2p0-legal/chat/completions \
  -H "Content-Type: application/json" \
  -H "apikey: YOUR_IAPP_API_KEY" \
  -d '{
  "model": "openthai2.0-legal",
  "rag": true,
  "rag_inject": "system",
  "messages": [
    {
      "role": "system",
      "content": "You are a Thai legal expert. Answer with legal analysis and cite the relevant มาตรา."
    },
    {
      "role": "user",
      "content": "จำเลยขีดฆ่าและฉีกเอกสารหลักฐานแห่งหนี้ แม้ยังอ่านได้ ถือเป็นความผิดสำเร็จหรือเพียงพยายามกระทำผิด"
    }
  ],
  "temperature": 0,
  "top_p": 1,
  "max_tokens": 4096,
  "chat_template_kwargs": {
    "enable_thinking": true
  },
  "stream": true,
  "stream_options": {
    "include_usage": true
  }
}'

Measured​

Citation rows use the NitiBench citation-F1 scorer (single pass, temperature 0, thinking off); holding, coverage and fluency in the essays are judged by an LLM. Every model was re-run under one protocol; the scoreboard shows every axis.

Run it​

from openai import OpenAI
client = OpenAI(base_url="https://api.iapp.co.th/v3/llm/openthai2p0-legal", api_key="YOUR_IAPP_API_KEY")

r = client.chat.completions.create(
    model="openthai2.0-legal",
    messages=[{"role": "user", "content": "ลักทรัพย์ในเวลากลางคืน ผิดมาตราใด"}],
    max_tokens=1024,
    # extra_body={"rag": False}  # uncomment to call the bare model (closed-book)
)
print(r.choices[0].message.content)
print(r.model_extra.get("retrieved_documents"))  # the law sections the answer used

curl -s https://api.iapp.co.th/v3/llm/openthai2p0-legal/chat/completions \
  -H "Content-Type: application/json" -H "apikey: YOUR_IAPP_API_KEY" \
  -d '{"model":"openthai2.0-legal","messages":[{"role":"user","content":"ลักทรัพย์ในเวลากลางคืน ผิดมาตราใด"}],"max_tokens":1024}'

vllm serve iapp/openthai2.0-legal-thaillm-nemotron-3-nano-30b-a3b \
  --tensor-parallel-size 2 --trust-remote-code \
  --max-model-len 32768 --enforce-eager

docker run --rm --gpus all --shm-size=16g \
  -v <path-to-model>:/model \
  -e NIM_MODEL_NAME=/model \
  -e NIM_SERVED_MODEL_NAME=openthai2.0-legal \
  -p 8000:8000 \
  nvcr.io/nim/nvidia/llm-nim:latest

Settings that matter​

"chat_template_kwargs": {"enable_thinking": true} returns the reasoning inside <think>...</think>. Serve with --max-model-len at 32768 or more: real Revenue Code contexts overflow smaller windows.

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")

SYSTEM = ("You are OpenThaiGPT-Legal, an expert assistant on Thai law. You are given a legal "
          "question and the exact statutory sections needed to answer it. Reason step by step in "
          "English, then give the final answer in Thai. Cite ONLY sections present in the provided "
          "context, using each section's exact law_name and bare section number (e.g. 132, 77/1). "
          'Output the final answer as JSON: {"answer": "<Thai answer>", '
          '"citations": [{"law": "<law_name>", "section": "<bare id>"}]}.')

USER = """Provided context:
<law law_name="ประมวลกฎหมายแพ่งและพาณิชย์" section="420">
ผู้ใดจงใจหรือประมาทเลินเล่อ ทำต่อบุคคลอื่นโดยผิดกฎหมายให้เขาเสียหาย ... จำต้องใช้ค่าสินไหมทดแทนเพื่อการนั้น
</law>

Question (ตอบเป็นภาษาไทย):
นาย ก. ขับรถโดยประมาทชนรถของนาย ข. เสียหาย นาย ก. ต้องรับผิดตามกฎหมายใด"""

r = client.chat.completions.create(
    model="openthai2.0-legal-thaillm-nemotron-3-nano-30b-a3b",
    messages=[{"role": "system", "content": SYSTEM}, {"role": "user", "content": USER}],
    temperature=0.0, max_tokens=1024,
    extra_body={"chat_template_kwargs": {"enable_thinking": False}})
print(r.choices[0].message.content)

{"answer": "นาย ก. ต้องรับผิดตามประมวลกฎหมายแพ่งและพาณิชย์ มาตรา 420",
 "citations": [{"law": "ประมวลกฎหมายแพ่งและพาณิชย์", "section": "420"}]}

SYSTEM = ("You are an expert on Thai law. You are given ONLY a legal question, with NO reference "
          "material provided. Using your OWN knowledge of Thai statutes, answer in Thai and cite the "
          "specific sections that apply (law name + bare section number, มาตรา). "
          'Output ONLY a JSON object: {"answer":"<Thai answer>","citations":[{"law":"<law name>",'
          '"section":"<bare section number e.g. 40 or 77/1>"}]}.')

SYSTEM = "You are a Thai legal expert. Answer the question with legal analysis and cite the relevant มาตรา."

To add your own retrieval, follow the Open WebUI and OpenThaiRAG tutorial and chunk the statutes one มาตรา per chunk with law_name and section metadata. To deploy it in your organisation or tune it to your own corpus, write to sale@iapp.co.th or call 02-124-4041.

Limits​

• Decision support, not legal advice: a qualified professional checks every citation against the current law.
• The answer follows the sections it is given, so wrong retrieval gives a wrong answer. With the law in the prompt, choosing the applicable section among close alternatives is the lowest-scoring test for every model.
• Trained on Thai statutes (Revenue Code, Civil and Commercial Code, related acts) and Revenue Department rulings; niche areas and recent amendments are thinner.

Licence​

NVIDIA Open Model Agreement. Trained from the text core of NVIDIA Nemotron-3-Nano-Omni-30B-A3B-Reasoning: continued pretraining on 360,985 statute drills, fine-tuning on 16,436 exam answers and essays, then reinforcement learning on 8,568 graded questions with citation F1 as the reward. The launch story has the detail.

Some essay training data derives from Thai bar examination materials; confirm redistribution terms for your use. Evaluated on NitiBench (VISAI-AI, MIT); fine-tuning data built from WangchanX-Legal-ThaiCCL (VISTEC, MIT). Built by the OpenThai team (AIEAT and iApp Technology) with the Big Data Institute, the ThaiLLM initiative and NVIDIA.