Our Latest Models
Typhoon 2.1 Gemma
Latest Released Model
Our newest release, Typhoon 2.1, introduces lightweight, high-performance models at 4B and 12B sizes—designed to rival larger models while remaining deployable on CPUs, L4s, and even consumer GPUs. It features improved Thai alignment, controllable thinking mode for long-form reasoning, and enhanced code-switching performance for real-world Thai-English use cases.
Typhoon 2
Typhoon 2 features a comprehensive model lineup spanning from lightweight edge-optimized variants (1B and 3B parameters) to a powerful 70B parameter model, all specifically engineered for Thai language applications. The architecture demonstrates exceptional benchmark performance while supporting extensive context lengths up to 128,000 tokens, incorporating robust safety classification systems and sophisticated function-calling capabilities to enable advanced AI applications.
Typhoon 2 R1 (Preview)
Built on the solid foundations of Typhoon 2 and Deepseek R1, Typhoon2-R1-70B delivers sharper reasoning and improved Thai language generation. Whether it's solving complex math, writing code, or navigating billingual tasks, this model brings balanced performance and versatility to the table
Frequently Asked Questions (FAQs)
What is Typhoon?
Typhoon is an open-source AI research initiative focused on creating advanced large language models (LLMs) specifically optimized for the Thai language. We aim to enhance cultural understanding, preserve Thai language intricacies, and empower the local tech community with cutting-edge AI tools.
How is Typhoon different from other AI models?
Typhoon models are uniquely tailored to the Thai language and its nuances. They are built with efficiency in mind, offering performance that is more cost-effective than other major models. Additionally, Typhoon is committed to open-source development, providing free access to model weights, APIs, datasets, and research to foster community-driven innovation.
What are the goals of Typhoon?
Typhoon aims to be recognized as one of the leading AI research labs in Southeast Asia and globally. Our focus is on developing state-of-the-art Thai language models and multimodal AI systems, all while maintaining our commitment to open-source principles and impactful research for the public good.
What is included in the Typhoon Series?
The Typhoon series includes several key components:
-
Typhoon LLMs: A range of open-source language models optimized for Thai, including 8B and 70B model sizes, built using frameworks like LLaMA 3 and Qwen.
-
Typhoon Audio (Research Preview): Advanced audio-language models capable of transcription, speech translation, Q&A, and audio captioning.
-
Typhoon Vision (Research Preview): Multimodal models integrating visual understanding with Thai language, designed for image input and interpretation.
How can I access Typhoon models?
Typhoon’s models are available open-source. You can access model weights on Hugging Face. Development APIs, datasets, and tools are available through our platform, https://opentyphoon.ai/. We believe in fostering collaboration, so all our research outputs are freely available for developers, researchers, and academics to use and build upon.
Is there a production-grade, paid, per-token billed Typhoon API?
Yes! We've partnered with Together AI to offer Typhoon API Pro – a paid, production-grade API through the Together Serverless platform. It is billed at Typhoon 2 70B: $0.88/1M tokens and Typhoon 2 8B: $0.18/1M tokens (login required).
What are the ongoing research areas for Typhoon?
Our ongoing research includes improving Thai language models, enhancing multimodal capabilities (such as integrating visual and audio inputs), and building solutions that preserve and reflect Thai cultural nuances. We are also exploring new applications in various sectors to make AI tools more accessible and impactful across different industries.
TYPHOON