ThaiLLM Gives Thailand a Free National AI Model
Four Thai teams, including SCB 10X and KBTG, built the models on a state supercomputer, and developers can call them through an OpenAI-style API.
Thailand's state science agency has released its own Thai-language AI model, ThaiLLM, in two sizes of 8 billion and 30 billion parameters, trained on more than 100 billion tokens of Thai text. The Nation Thailand reported the launch on 20 April 2026, and the model runs on a domestic supercomputer so that data can be stored and processed inside the country.
Parameters are the adjustable numbers inside a model. More of them usually means a more capable model that also costs more to run. A token is a small chunk of text, so 100 billion tokens is a very large reading list.
Four teams, one release
ThaiLLM is not a single model built by a single lab. The National Science and Technology Development Agency (NSTDA) led it with the Ministry of Higher Education, Science, Research and Innovation (MHESI), its NECTEC electronics centre and the Digital Economy and Society Development Fund. More than 700 people took part, according to the Nation.
The first phase packages four 8 billion parameter models from different groups:
- OpenThaiGPT, from AIEAT
- Pathumma, from NECTEC
- Typhoon-S, from SCB 10X
- THaLLE, from KBTG
Two of those names come from the private sector. KBTG is the KASIKORN Business-Technology Group. Both KBTG and SCB 10X are named as companies working with the models, alongside finance, technology and healthcare organisations, and VISTEC is running experimental medical work.
What a developer actually gets
Access is through an API that is compatible with the OpenAI software kit, so a team that has already written code against that interface can point it at ThaiLLM with small changes. There is also a public chat page, the ThaiLLM Playground, which shows its answers with cited sources.
The Ministry of Digital Economy and Society (DE) held the formal launch on 5 May 2026, according to a briefing published by Thailand Edition. It described the 8B and 30B models as open-licence and open-weight, meaning developers can download and fine-tune them, and it said healthcare is the first focus. Over 8,000 people had used the Playground since early 2026. NT and SIAM.AI CLOUD supply infrastructure for the Playground.
A deputy permanent secretary at MHESI said ThaiLLM is meant to move Thailand "from user to creator" of AI. The DE permanent secretary said the point is to understand the language as used in Thai social context, not only to access tools.
The Nation describes ThaiLLM as national AI infrastructure. Thailand NOW confirms the same builders and sizes. None of the sources reviewed for this story publish benchmark scores against those models, so how ThaiLLM compares on accuracy is not yet public.
The money question
No source gives a budget for ThaiLLM. The launch briefing says the Digital Economy and Society Development Fund provided funding but gives no amount.
The nearest public figure is a separate one. In August 2025 Thailand's National AI Committee endorsed a THB 25 billion framework for two fiscal years to 2027, Thailand Business News reported: THB 6 billion for AI workforce development, THB 5 billion for building and running Centres of Excellence and THB 2 billion for a National Data Bank. One of its nine planned Centres of Excellence covers language processing. That article does not mention ThaiLLM, so the two should not be added together.
Rules are catching up
The policy frame is also moving. The National AI Committee was formed in 2025 under the Prime Minister with the aim of making Thailand an ASEAN AI hub by 2027, according to the Tech For Good Institute. A draft Thailand AI Act went to public consultation from 9 July to 14 August 2026. It would give the Electronic Transactions Development Agency (ETDA) powers to order corrective action or block an AI system, and it requires disclosure of AI-generated content.
For a Thai company that handles customer data, the practical draw is simple: a Thai-tuned model that can be hosted under Thai jurisdiction. The open question is quality. The thing to watch is the outcome of the AI Act consultation and whether the Playground user count grows past the 8,000 reported in May.
FAQ
What is ThaiLLM?
It is a Thai-language large language model released in 2026 by NSTDA, MHESI, NECTEC and the Digital Economy and Society Development Fund. It comes in 8 billion and 30 billion parameter sizes.
Who built the models inside ThaiLLM?
The first phase includes four 8B models, OpenThaiGPT from AIEAT, Pathumma from NECTEC, Typhoon-S from SCB 10X and THaLLE from KBTG.
Can a business use ThaiLLM today?
Developers can reach it through an API compatible with the OpenAI software kit and through the ThaiLLM Playground. The launch briefing described the models as open-licence and open-weight.
How much did ThaiLLM cost?
No source reviewed gives a figure. The Digital Economy and Society Development Fund provided funding, but no amount has been published.
Sources
- Nation Thailand: ThaiLLM launched to strengthen Thailand's national AI capacity
- NSTDA: ThaiLLM, Thailand's national Thai-language model
- Thailand Edition: DE-MOHE Launches ThaiLLM
- Thailand NOW: ThaiLLM, what you need to know about Thailand's homegrown AI project
- Tech For Good Institute: Governance Meets Reality, Thailand Tech Governance at mid-2026
- Thailand Business News: Thailand Commits 25 Billion Baht to Drive AI Development Through 2027
