News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
fnTyphoon

Typhoon

Thai-focused large language model series by SCB 10X

Typhoon is a series of large language models built for the Thai language, developed by SCB 10X. The latest version, Typhoon2, uses the Llama 3.1 architecture with 70 billion parameters to handle Thai and English tasks, including coding and math.

Share on WhatsApp

  • Based on Llama 3.1 with 70 billion parameters
  • Optimized for Thai language and code-switching
  • Supports function calling and tool use
  • Released under Llama 3.1 Community License
  • Outperforms previous open-source Thai models

What it is

Typhoon is a family of large language models designed specifically for the Thai language. Developed by SCB 10X, the series aims to address the lack of high-quality open-source models for Thai speakers. The most recent iteration, Typhoon2, is built on the Llama 3.1 foundation and contains 70 billion parameters. It supports both Thai and English, with a focus on instruction following, mathematical reasoning, and coding tasks. The model is intended for developers and researchers who need robust performance in Thai contexts.

How it came about

The project began with an initial technical report in December 2023, which introduced the first Typhoon models. These early versions used continual training to transfer knowledge from existing large models to improve Thai language capabilities. The team evaluated these models using ThaiExam, a benchmark based on local high school and professional exams. In December 2024, the team released Typhoon2, scaling up to 70 billion parameters and aligning it with the Llama 3.1 architecture to enhance performance across multiple domains.

Inside the model

Typhoon2 is a decoder-only transformer model based on the Llama 3.1 architecture. It features 70 billion parameters and is optimized for instruction following. The model supports function calling, allowing it to interact with external tools such as weather APIs or stock price lookups. Training involved adapting the base Llama model to better handle Thai text, which is tokenized more efficiently than in many other models. The system requires transformers version 4.45.0 or newer and can be served using vLLM with FP8 quantization for faster inference.

What it can do

The model demonstrates strong performance in Thai language tasks. On the IFEval benchmark for Thai, it scored 81.45 per cent, outperforming some larger general models in specific Thai metrics. It handles code-switching between Thai and English effectively, achieving 98.8 per cent accuracy at a temperature of 0.7. In mathematical reasoning, it scored 88.79 per cent on GSM8K in Thai. While its English performance is competitive, it shows particular strength in Thai-specific coding and function calling tasks compared to other open-source alternatives.

How to use it

Typhoon2 is available under the Llama 3.1 Community License. The weights are hosted on Hugging Face under the identifier scb10x/llama3.1-typhoon2-70b-instruct. Developers can download the model for local use or run it via API services like opentyphoon.ai. Running the full 70 billion parameter model requires significant hardware, such as two 80GB GPUs (A100 or H100), though quantization options allow for more efficient deployment.

Safety and testing

The model includes guardrails to prevent harmful outputs, but the developers note it is still under development. It may produce inaccurate or biased responses. Users are advised to assess risks for their specific use cases. The system prompt instructs the assistant to be helpful, harmless, and honest, avoiding unnecessary filler phrases. It is designed to respond directly in the language requested by the user.

Why it matters

Typhoon has become a key reference for Thai language processing in the open-source community. By providing a high-performance model tailored to Thai, it enables local developers to build applications without relying solely on closed, English-centric systems. The release of Typhoon2, with its integration into the Llama ecosystem, further standardizes high-quality Thai language support for enterprise and research applications.

Questions people ask about Typhoon

What is Typhoon?

Typhoon is a series of large language models developed by SCB 10X specifically for the Thai language. It helps developers build AI applications that understand and generate Thai text effectively.

How many parameters does Typhoon2 have?

The Typhoon2 model, specifically the Llama3.1-Typhoon2-70B-instruct version, has 70 billion parameters. It is based on the Llama 3.1 architecture.

Can Typhoon handle English as well?

Yes, Typhoon is designed to handle both Thai and English. It performs well in code-switching scenarios and maintains strong capabilities in English tasks like coding and math.

Is Typhoon free to use?

The model weights are available under the Llama 3.1 Community License. This allows for broad use, but developers must comply with the license terms set by Meta and SCB 10X.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Typhoon yet, so the facts on this page come from the sources listed above: huggingface.co/scb10x/llama3.1-typhoon2-70b-instruct, arxiv.org/abs/2312.13951. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.