Model reference · open weights

openthaigpt1.5

LLMs openthaigpt Text gen 1 build Its own licence terms 83k dl/mo

openthaigpt1.5 is an open-weight language model from openthaigpt. openthaigpt1.5-7b-instruct (BF16) weighs 15.2 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.

For self-hosting, compare this build's requirements below with the GPU you plan to rent.

  • OpenThaiGPT 1.5 is a 7.6B parameter text-generation model developed by openthaigpt for Thai and English language tasks.
  • It supports a context length of 32,768 tokens and is fine-tuned on over 2,000,000 Thai instruction pairs.
  • The model is released under a license that permits research and commercial use, with specific conditions applying if the user base exceeds 100 million monthly active users.

Summary of the openthaigpt/openthaigpt1.5-7b-instruct model card, 2026-10-01

What it is

Released byopenthaigpt
Released2024-09-30
Parameters7.6B
VRAM15.2 GB for the weights

What it runs on

Memory and cards for openthaigpt1.5-7b-instruct (BF16)

15.2 GBweights, file size
57 MBcache per 1K tokens
702 MBruntime overhead, at least
32,768 tokenscontext max
CardRequests at onceContext maxMemory
8K each32K each
RTX 3060 12 GB … RTX 4060 Ti 16 GB
2 smaller cards
RTX 3090 24 GB153all 32K23.4 GB
RTX 4090 24 GB153all 32K23.4 GB
RTX 5090 32 GB328all 32K31.0 GB
L40S 48 GB5914all 32K44.0 GB
A100 80 GB13233all 32K78.2 GB
H100 80 GB13233all 32K78.1 GB
RTX PRO 6000 Blackwell 96 GB16541all 32K93.8 GB
DGX Spark (GB10) 128 GB unified19448all 32K107 GB
H200 141 GB25964all 32K138 GB
B200 180 GB34185all 32K176 GB
2× RTX 3060 12 GB
split by layers (llama.cpp)
143all 32K11.6 GB a card
2× RTX 4060 Ti 16 GB
split by layers (llama.cpp)
307all 32K15.4 GB a card
Memory needed at each load
Requests at once8K tokens each32K tokens each
116.4 GB17.8 GB
518.3 GB25.3 GB
819.7 GB31.0 GB
1623.5 GB46.0 GB
3231.0 GB76.1 GB
6446.0 GB136 GB

One card, with vLLM's small-card settings.

From the model card

What openthaigpt says about openthaigpt1.5

Read the model card

More Info

🇹🇭 OpenThaiGPT 7b Version 1.5 is an advanced 7-billion-parameter Thai language chat model based on Qwen v2.5 released on September 30, 2024. It has been specifically fine-tuned on over 2,000,000 Thai instruction pairs and is capable of answering Thai-specific domain questions.

Online Demo:

https://demo72b.aieat.or.th/

Example code for API Calling

https://github.com/OpenThaiGPT/openthaigpt1.5_api_examples

📢 OpenThaiGPT is now OpenThai

The project was renamed in July 2026. Released models keep their original names, and this repository path is unchanged and permanent. → openthai.aieat.or.th

Other models in the family

NeedModel
⚖️ Thai law, statute citation, legal RAGOpenThai 2.0 Legal 30B-A3B ← newest
🧠 Reasoning, maths, logic, codeOpenThaiGPT R1 32B
💬 General Thai chat and codingOpenThaiGPT 1.6 72B
💻 Limited GPUOpenThaiGPT 1.5 7B

⚡ Run it locally

ollama run openthai/openthai-1.5-7b

Official GGUF quants (Q4_K_M / Q5_K_M / Q8_0): openthaigpt/openthaigpt1.5-7b-instruct-GGUF — runs in ~8 GB RAM, no GPU needed. Step-by-step guide: openthai.aieat.or.th/ollama

Highlights

  • State-of-the-art Thai language LLM, achieving the highest average scores across various Thai language exams compared to other open-source Thai LLMs.
  • Multi-turn conversation support for extended dialogues.
  • Retrieval Augmented Generation (RAG) compatibility for enhanced response generation.
  • Impressive context handling: Processes up to 131,072 tokens of input and generates up to 8,192 tokens, enabling detailed and complex interactions.
  • Tool calling support: Enables users to efficiently call various functions through intelligent responses.

Benchmark on OpenThaiGPT Eval

** Please take a look at openthaigpt/openthaigpt1.5-7b-instruct for this model's evaluation result.

Exam namesscb10x/llama-3-typhoon-v1.5x-8b-instructmeta-llama/Llama-3.1-7B-InstructQwen/Qwen2.5-7B-Instruct_statopenthaigpt/openthaigpt1.5-7b
01_a_level46.67%47.50%58.33%60.00%
02_tgat32.00%36.00%32.00%36.00%
03_tpat152.50%55.00%57.50%57.50%
04_investment_consult56.00%48.00%68.00%76.00%
05_facebook_beleble_th_20078.00%73.00%79.00%81.00%
06_xcopa_th_20079.50%69.00%80.50%81.00%
07_xnli2.0_th_20056.50%55.00%53.00%54.50%
08_onet_m3_thai48.00%32.00%72.00%64.00%
09_onet_m3_social75.00%50.00%90.00%80.00%
10_onet_m3_math25.00%18.75%31.25%31.25%
11_onet_m3_science46.15%42.31%46.15%46.15%
12_onet_m3_english70.00%76.67%86.67%83.33%
13_onet_m6_thai47.69%29.23%46.15%53.85%
14_onet_m6_math29.41%17.65%29.41%29.41%
15_onet_m6_social50.91%43.64%56.36%58.18%
16_onet_m6_science42.86%32.14%57.14%57.14%
17_onet_m6_english65.38%71.15%

Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

Benchmarks

Reported results

As published on the model card: the maker's own numbers, not measured by AxForge.

TaskDatasetMetricScore
text-generationThaiExamThai Exam(Acc)52.040
text-generationM3ExamM3Exam(Acc)54.010
© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms