- 5
- April
Have you ever wondered — "Why can ChatGPT respond in Thai, yet still struggles with Thai-specific topics?" Whether it is royal vocabulary, Thai idioms, Thai laws, or government regulations, foreign AI models often get things wrong or provide incomplete answers because their training data is predominantly English. ThaiLLM is the answer — a national AI infrastructure developed by Thai people, specifically for the Thai language and Thai context.
What Is ThaiLLM?
ThaiLLM (Thai Large Language Model) is a national-level AI infrastructure developed to give Thailand its own Thai-language AI, reducing dependence on foreign AI systems. It was trained on over 100 billion Thai-language tokens covering government documents, research papers, legislation, news, and various domain-specific Thai data.
ThaiLLM in 30 Seconds
- What it is: A national-level Thai-language AI developed by Thai agencies
- Trained on: 100+ billion Thai-language tokens
- Two versions: 8B parameters (fast) and 30B parameters (high performance)
- Free to use at: thaillm.or.th (Playground, API, Model Downloads)
- Key strength: Deep understanding of Thai language, idioms, laws, and government regulations
The core concept behind ThaiLLM is Digital Sovereignty — meaning Thailand can control its own data and AI technology without sending government or confidential organizational data to be processed on Big Tech cloud servers overseas. This is a critical data security concern that every organization must consider.
Who Is Behind ThaiLLM?
ThaiLLM was not created by a single agency. It is a national-level collaboration between government bodies, research institutions, universities, and the private sector:
| Organization | Role |
|---|---|
| NSTDA | Lead R&D agency; provides LANTA HPC computing resources |
| NECTEC | Develops Pathumma LLM, the core model of ThaiLLM supporting Text, Vision, and Audio |
| MHESI | Provides policy support and budget; promotes AI as national infrastructure |
| Siam AI Corporation | Co-develops models and drives commercial applications in the business sector |
| DEF Fund | Provides funding for developing AI for public use |
| BDI + Nation Group | Supports data provision and expansion into media and business sectors |
In addition, several leading universities participate in developing specialized sub-models and training datasets for specific domains such as medicine, law, and agriculture.
Pathumma LLM — 3 Core Capabilities
Pathumma LLM is the core model developed by NECTEC under the ThaiLLM project. The name "Pathumma" comes from the lotus flower, a symbol of Thailand. This model does not just read and write Thai — it covers three comprehensive capabilities:
| Capability | Description | Example Use Cases |
|---|---|---|
| Text LLM | Processes Thai text — writing, summarizing, translating, and analyzing | Summarizing government documents, drafting official letters, analyzing legislation |
| Vision LLM | Reads and understands images, documents, and charts in Thai | Reading receipts, analyzing graphs, reading Thai signage |
| Audio LLM | Listens to and understands spoken Thai, transcribing speech to text | Transcribing meetings, recording instructions, converting speech to documents |
All three capabilities work together, making Pathumma a truly multimodal AI. For example, you can photograph a document and ask questions about it in Thai, or submit an audio file and have it summarized as text.
ThaiLLM vs ChatGPT vs Gemini — A Direct Comparison
A common question is "Is ThaiLLM better than ChatGPT or Gemini?" The answer is: it is not better at everything, but it excels in contexts that matter most to Thai users. Here is a comparison:
| Criteria | ThaiLLM | ChatGPT / Gemini |
|---|---|---|
| Thai Language Understanding | Excellent — trained specifically on Thai data | Good — but Thai is a small portion of training data |
| Thai Government/Legal Data | Comprehensive — trained on real government documents | Limited — often inaccurate or outdated |
| Cost | Free (Playground + API) | Limited free tier / paid subscription |
| Data Control | Data stays in Thailand (Sovereign AI) | Data sent to overseas cloud servers |
| Model Downloads | Yes — run on your own organization's servers | No — must use the provider's API only |
| General Capabilities (English, Coding) | Moderate | Excellent |
Key Takeaway: ThaiLLM is not meant to replace ChatGPT or Gemini. It is a critical alternative for tasks requiring deep Thai language understanding and for work that demands in-country data control — especially for government agencies and organizations handling confidential business data.
How to Use ThaiLLM — 3 Channels
ThaiLLM is available for free through three main channels:
| Channel | Best For | How to Get Started |
|---|---|---|
| Playground | General experimentation — type and get answers instantly | Visit thaillm.or.th → Select Playground → Type your question |
| API | Developers / integrating into organizational systems | Register for an API Key → Call via REST API |
| Model Downloads | Organizations that need to run the model on their own servers | Download the model → Deploy on an in-house GPU server |
Benefits of Running the Model On-Premise
- Data never leaves the organization — ideal for government secrets and business-critical data
- No usage limits (no rate limiting)
- Fine-tune the model to fit your organization's specific tasks
- No internet dependency — works even on internal networks
Bootcamp LLM Research Challenge Thailand 2026
One key initiative demonstrating that ThaiLLM is not just "build a model and leave it" is the Bootcamp LLM Research Challenge Thailand 2026 — a training program that equips government IT personnel with the ability to apply AI to solve real problems in their agencies:
| Detail | Figure |
|---|---|
| Government IT personnel participating | 110+ from multiple ministries and departments |
| Innovation prototypes created | 24 prototypes covering various government missions |
| 1st Prize | Team B05 — BAAC (Bank for Agriculture and Agricultural Cooperatives) (60,000 THB) |
| 2nd Prize | Team B01 — Public Relations Department (40,000 THB) |
The winning BAAC team used ThaiLLM to analyze farmer and loan data — a prime example of how Thai-language AI can deliver real benefits at the operational level. This initiative proves that ThaiLLM is not a "toy" but a production-ready tool for government agencies.
Why Sovereign AI? — Digital Sovereignty Explained
The term "Sovereign AI" may sound technical, but its meaning is straightforward — Thai data should remain in Thai hands.
Today, when we use ChatGPT or Gemini, the data we type is sent to overseas servers. For personal use, this may not be a concern. But for government agencies, financial institutions, or organizations handling confidential data, this is an unacceptable risk. Once data leaves the organization, there is no way to know how it will be used. This aligns with the principle of data security that "the safest data is data that never leaves your premises."
ThaiLLM solves this because:
- Data is processed within Thailand — never sent abroad
- The model can be downloaded — run on your own organization's servers
- Reduces dependence on Big Tech — no fear of price increases or policy changes
- Fully customizable — fine-tune the model to address your organization's specific needs
ThaiLLM and ERP Systems — New Opportunities for Thai Organizations
A Thai-language AI like ThaiLLM opens new possibilities for integrating AI into ERP systems. Historically, using AI with ERP has been hampered by language barriers — documents are in Thai, instructions are in Thai, but the available AI primarily understands English.
Saeree ERP is currently developing an AI Assistant to help users work faster. One approach under evaluation is applying Thai-language models. Here are some potential use cases:
| Use Case | Description | Benefit |
|---|---|---|
| Purchase Order Analysis | AI reads Thai-language purchase orders, verifying items, prices, and conditions | Reduces document review time and errors |
| Report Summarization | AI summarizes budget and inventory reports into easy-to-read Thai | Executives understand data faster, enabling immediate decisions |
| User Q&A | AI answers questions about the ERP system in Thai | Reduces IT team workload; users solve problems independently |
| Document Classification | AI reads Thai documents and automatically categorizes them | Faster document retrieval, reduced redundant work |
Having a Thai-language AI that understands Thai context dramatically improves AI integration with ERP systems. There is no longer a need to "translate Thai to English" before sending it to AI and then "translate back to Thai" — a step where errors commonly occur. For organizations interested in learning why Saeree ERP, more details are available.
Furthermore, having a well-organized Data Warehouse provides the essential foundation for AI to analyze data more accurately — because AI is only as smart as the data it receives.
A Thai-language AI that truly understands Thai context — not just translating from English — represents a significant step toward Digital Sovereignty that will transform how both government and private organizations in Thailand operate.
- ThaiLLM Research Team
Summary — Why ThaiLLM Matters
- ThaiLLM is a national-level Thai AI developed by NSTDA + NECTEC + partners, trained on 100+ billion Thai tokens
- Pathumma LLM offers 3 capabilities — Text, Vision, and Audio covering a wide range of tasks
- Free to use at thaillm.or.th — includes Playground, API, and downloadable models for self-hosting
- Outperforms ChatGPT/Gemini in Thai contexts — government data, legislation, Thai idioms
- Supports Digital Sovereignty — data stays in Thailand, reducing dependence on Big Tech
- Production-ready — Bootcamp 2026 produced 24 prototypes from 110+ government personnel
- An opportunity for ERP — Thai-language AI will make ERP systems smarter within the Thai context
References
- National Science and Technology Development Agency (NSTDA)
- National Electronics and Computer Technology Center (NECTEC)
- ThaiLLM — Playground, API, Model Downloads
- ai.in.th — AI Platform for Thai People
If your organization is looking for an ERP system ready to support future AI technologies, you can schedule a demo or contact our consulting team for an organizational readiness assessment.
