WangchanX
VISTEC's open Thai fine-tuning pipeline, models and instruction datasets
Overview
WangchanX is the open Thai language-model project of the VISTEC-depa AI Research Institute of Thailand, whose listed supporters include AIS, IRPC, PTT Global Chemical and Siam Commercial Bank. It covers an Apache 2.0 fine-tuning pipeline (SFT, DPO, ORPO), the WangchanLion model built with AI Singapore, the human-written WangchanThaiInstruct dataset and the WangchanBERTa encoder. Everything is self-hosted; there is no API or paid service.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 19 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- SFT, DPO and ORPO fine-tuning scripts (Apache 2.0)
- WangchanLion and WangchanX fine-tuned model weights
- WangchanThaiInstruct and a 120k synthetic instruction set
- Deployment notes for TGI, LocalAI and Ollama
- No API or support contract from the institute
Use cases
What you can produce with WangchanX
- A Thai instruction-tuned model fine-tuned on your own data with the WangchanX SFT, DPO or ORPO scripts
- A self-hosted Thai chatbot served through TGI, LocalAI or Ollama, with no data leaving your servers
- A Thai sentiment classifier or named-entity tagger built on WangchanBERTa
- A Thai legal question-answering setup using the WangchanX legal retriever and ThaiCCL RAG dataset
- An evaluation of Thai LLMs on the WangchanThaiInstruct test split across four domains
- Reading-comprehension benchmark scores from the WangchanX-Eval code
ASEAN Perspective
WangchanX in Southeast Asia
WangchanX is Thailand's most prominent nationally-funded open LLM initiative, developed inside the VISTEC-depa AI Research Institute — a government-backed (depa) research body. Its collaboration with AI Singapore on the WangchanLION / SEA-LION line makes it a concrete bridge between Thai and wider ASEAN NLP ecosystems, supporting the region's push for sovereign, locally-hosted AI. Because all weights are self-hostable under Apache 2.0, Thai organisations in regulated sectors (government, healthcare, finance) can run models entirely within Thai data-centre boundaries, satisfying PDPA data-residency expectations without reliance on US or Chinese cloud APIs. The WangchanX-Legal-ThaiCCL-RAG dataset also demonstrates domain relevance to ASEAN legal systems, which share some structural similarities with Thai corporate and commercial law.
WangchanX is useful infrastructure for Thai-language work: a documented pipeline for SFT, DPO and ORPO, a well-used Thai encoder in WangchanBERTa, and WangchanThaiInstruct, a 40,000-example human-written dataset with a peer-reviewed paper behind it. For teams that need data to stay on their own servers, it covers more of the workflow openly than most Thai projects.
It is not a product. The pipeline has been quiet since October 2024, the WangchanX-branded models stop at 13B, licences vary by component, and Thailand's Qwen3-based ThaiLLM models and Typhoon 2.5 now set the pace. Best for researchers and ML engineers, not for anyone wanting a hosted service.
What we found
WangchanX is the name the VISTEC-depa AI Research Institute of Thailand uses for its open Thai language-model work: a fine-tuning pipeline, instruction-tuned models, datasets and evaluation code, released through GitHub and Hugging Face rather than sold. The pipeline, Apache 2.0 since its April 2024 release, runs supervised fine-tuning, DPO and ORPO on bases such as Llama 3, PolyLM, SEA-LION and Typhoon, with deployment notes for TGI, LocalAI and Ollama. Its last commit was in October 2024. Legal work followed in 2025: a retriever and a question-answering dataset for Thai corporate and commercial law.
The pieces that kept their value span several years. WangchanBERTa, the institute's 2021 Thai encoder, logged about 143,000 Hugging Face downloads in the month to 19 October 2026, far more than anything else it publishes. WangchanThaiInstruct, a human-written instruction set covering medical, legal, finance and retail tasks, holds 40,000 examples, 35,014 of them in the subset used for its EMNLP 2025 paper. WangchanLion, a 7B model fine-tuned with AI Singapore on SEA-LION, posts the best zero-shot F1 on Thai XQuAD in VISTEC's own table, while Typhoon-7B leads in the one-shot setting.
The field has moved on. Thailand's shared ThaiLLM base models, built on Qwen3, now carry fine-tunes from SCB 10X's Typhoon, KBTG, AIEAT and NECTEC, and Typhoon itself has reached version 2.5. VISTEC's newer public repositories cover ThaiLLM medical post-training and patient pre-screening, plus Qwen3 30B-class research checkpoints, rather than new WangchanX-branded releases.
Nothing here is a product. There is no hosted API, chat app or support contract, and the institute's own website has not listed a new release since 2021. Licences need checking piece by piece: the pipeline and WangchanLion are Apache 2.0, the Llama 3 fine-tune follows Meta's licence, the 120,000-example synthetic set is MIT, and WangchanThaiInstruct mixes non-commercial and share-alike rows. For Thai teams that must self-host, it remains a documented starting point.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- The name 'Wangchan' comes from Wangchan Valley in Rayong — the science and technology zone where VISTEC's campus is physically located.
- WangchanBERTa was trained on 381 million unique Thai sentences (78.5 GB) using 8 NVIDIA V100 GPUs and has since spawned 58 community fine-tuned variants.
- WangchanLion is a joint Thai–Singaporean effort, built on AI Singapore's SEA-LION base model and then instruction-tuned specifically for Thai — demonstrating cross-border ASEAN AI collaboration.
- The WangchanThaiInstruct dataset (75K human-annotated examples, EMNLP 2025) is among the largest human-authored Thai instruction datasets ever released, covering medical, legal, finance, and retail domains.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including WangchanX's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with WangchanX unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to WangchanX directly →
Spotted something out of date? Suggest an update →
Related reading
Health & Wellness
Regular Tea Linked to Lower Heart Risk in China Study, but Burning-Hot Cups Went the Other...
Science
Eighteen Years of Whale Calls Show Rare Omura's Whales Favour Warm, Food-Rich Waters off A...
Health & Wellness
Fine-Particle Air Pollution Linked to Higher Hospital Admission Risk for Up to 21 Days, St...
Alternatives to WangchanX
More in LLMs & Chat
