ThaiLLM
Thailand's free, open-weight national LLM for Thai law and context
Overview
ThaiLLM is Thailand's government-backed Thai-language model, launched April 2026 by MHESI/NSTDA and NECTEC on the domestic ThaiSC LANTA supercomputer. Ships as four 8B/30B sub-models with open weights and a free API, already piloted by KBTG, SCB 10X, and VISTEC.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Jul 2026 and may be out of date or incomplete.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.
What you can produce with ThaiLLM
- Open-weight 8B and 30B Thai-language models (four sub-models)
- Free public Playground and developer API
- Downloadable weights via Hugging Face for self-hosting
- Domestic-only data processing on ThaiSC LANTA supercomputer
- Trained on 100+ billion tokens of Thai-language data
ASEAN Perspective
ThaiLLM in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
ThaiLLM's real differentiator is what it was trained on and where it runs: Thai-language data processed entirely inside the country, which matters for anyone handling government or PDPA-sensitive data who doesn't want it leaving Thai borders. Early evaluation reports have it beating general frontier models on formal Thai writing and interpreting Thai legal and regulatory terminology — a narrow but genuinely useful edge.
It's not a frontier-model replacement. Launch version is text-only (no image or audio), the context window is well short of Claude or GPT-class models, and published rate limits — 5 requests/second, 200/minute — are workable for internal tools but not built for consumer-facing scale. Creative writing quality still trails the big commercial labs.
Who it's for: developers building Thai government, legal, or public-sector products who want open, self-hostable weights and a free API rather than routing sensitive Thai-language data through a foreign commercial provider. Best used alongside a frontier model, not instead of one.
What people say
ThaiLLM launched officially in April 2026 as a joint effort between Thailand's Ministry of Higher Education, Science, Research and Innovation (MHESI), NSTDA, and NECTEC, running on the domestic ThaiSC LANTA supercomputer so training and inference data stay within Thai borders. Rather than a single model, it bundles four sub-models from different Thai research and industry groups: OpenThaiGPT-ThaiLLM-8B-Instruct by AIEAT, Pathumma-ThaiLLM by NECTEC, Typhoon-S-ThaiLLM-8B by SCB 10X, and THaLLE-ThaiLLM-8B by KBTG, in 8B and 30B parameter sizes trained on 100+ billion tokens. Weights are open and downloadable from the ThaiLLM organization on Hugging Face, and there's a free public Playground plus a free developer API at launch.
Early hands-on assessment (via a Thai enterprise-tech analysis) found ThaiLLM's 8B model outperforming Claude Sonnet on Thai-context tasks — formal document writing, interpreting Thai laws and regulations, government terminology — in informal head-to-head testing, though without published accuracy percentages to back the claim precisely. The same assessment flagged real gaps: no multimodal support at launch, a context window well short of frontier commercial models, rate limits (5 req/sec, 200 req/min) suited to internal tooling rather than customer-facing apps at scale, weaker creative-writing output than frontier models, and no established enterprise SLA or uptime guarantee yet. The recommended use pattern is pairing ThaiLLM with a frontier model rather than replacing one outright.
On the research side, Thailand's broader Thai-LLM evaluation ecosystem (including the ThaiLLM Leaderboard and benchmarks like ThaiExam and ThaiCLI) is still young — researchers note existing Thai benchmarks lean heavily on machine-translated data and narrow, traditional NLP tasks like sentiment analysis and named-entity recognition, meaning claims of besting larger frontier models on Thai tasks specifically should be read against benchmarks that are themselves still maturing.
Adoption signals are early but concrete: KBTG (Kasikornbank's tech arm), SCB 10X (Siam Commercial Bank's venture/tech unit), and VISTEC are named as early adopters or contributing institutions, and the four-model bundle structure means each sub-model already has an existing user base from its originating organization.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including ThaiLLM's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with ThaiLLM 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 ThaiLLM directly →
Spotted something out of date? Suggest an update →
Alternatives to ThaiLLM
More in LLMs & Chat
