Typhoon vs OpenThai vs Pathumma: Three Thai LLMs Worth Knowing

Three Thai-language Large Language Models (LLMs) currently stand out for their unique strengths and open-source availability. Typhoon, developed by SCB 10X, boasts a comprehensive research portfolio with 30 million API calls, 2.6 million downloads, and 13,000 unique users. Its distinguishing feature is breadth, offering speech recognition, document reading, Thai-English translation, and reasoning models.

Notably, Typhoon Isan ASR pushes the boundaries by recognizing Isan speech, a dialect often overlooked by foreign models. OpenThai, led by the Thai AI Entrepreneur Association, offers Apache 2.0 licensed flagship models, allowing commercial use without fees. Its OpenThai 2.0 variant, with 27 billion parameters, excels in reading images, Thai handwriting, and performing agent-style tool calling.

For legal work, the OpenThai 2.0 Legal model, with 30 billion parameters, activates only 3 billion per token, running efficiently on a 24GB GPU. Pathumma, developed by a Thai research team at VISTEC, focuses on cultural and contextual understanding, a gap often present in foreign models. Unlike Typhoon and OpenThai, Pathumma prioritizes Thai cultural nuances, enabling it to better understand the subtleties of Thai language and law.

All three models run locally on adequate hardware, though their substantial size (27 billion parameters and up) demands significant GPU or RAM. Before choosing, carefully review each project's license to ensure compliance with commercial use. Although these models fall short of leading foreign models in multi-step reasoning, they excel in Thai-specific tasks and affordability. For those starting without installation, both OpenThai and Typhoon offer free web playgrounds.