As local AI development gains momentum in Thailand, many have heard of Typhoon, the open Thai-centric language model initiative—but not everyone fully understands its real value to enterprises, developers, and the broader ecosystem.
To help clarify this, Typhoon and our partners were invited to speak at Techsauce Global Summit, where interest around AI and local LLMs was higher than ever. The spotlight was on a special panel titled “AI Advantage for Thai Enterprises: Local Language Models in Action,” featuring experts from SCBX, SCB 10X, VISTEC, and TDRI including:
-
Kasima Tharnpipitchai, Head of AI Strategy, SCB 10X
-
Weerin Chantaroje, Head of Innovation Lab, SCBX
-
Narin Tananitaporn, Researcher, Thailand Development Research Institute (TDRI)
-
Dr. Sarana Nutanong, Head of the Natural Language Processing and Representation Learning Lab, VISTEC
In case you missed it or want to dive deeper, we’ve captured the four key insights of how to build AI advantages from local LLMs from this exclusive discussion as follows:
1. Unlocking Business Value: Tangible Benefits of Using Local LLMs
Weerin Chantaroje from SCBX shared firsthand experience on why local models can offer significant business advantages. From their experimentation building internal proof of concept using Typhoon and other models, SCBX found three key benefits:
1. Cost
Across multiple AI projects, SCBX has observed potential substantial cost savings by using local models developed by Typhoon. One example is a voice AI project, where switching to Typhoon cut the cost of serving a model by a remarkable 8 times compared to alternative solutions (Token costs by API offered from a propriatary model). This level of efficiency makes large-scale AI adoption across the business more sustainable.
2. Controllability
In October 2024, SCBX used a leading proprietary language model that initially performed well. However, when the provider rolled out a new version in December, the model’s performance in Thai unexpectedly degraded , and the team had no way to intervene.
The new model may perform better in other areas, but in that specific use case with Thai language, its performance declined. Such incidents could very well happen again, because the benchmarks used to evaluate foreign models may not include Thai.
You can’t let this kind of risk impact your business. We want reliable and controllable solutions
Weerin posed.
3. Customization
Businesses aiming for long-term, unfair advantage need more than off-the-shelf solutions. While global models provide strong baseline capabilities, they also provide the same experience to everyone. Customization, especially enabled by open-source models, lets organizations tailor AI to their exact needs, industry context, and workflows.
The best model is the right model for your business—one that balances cost, control, and customization.
2. Shift from ‘Silent Consumers’ to ‘Active Co-Creators’
Kasima Tharnpipitchai emphasized the importance of making Thai a “first-class citizen” in the world of AI. He highlighted how local models are crucial—not just for supporting the Thai language, but for capturing our culture, context, and nuance.
Without this local focus, Thai often falls outside the scope of performance benchmarks used by global models—leading to situations like the one Weerin described, where a model update caused a drop in Thai-language performance. When Thai isn’t prioritized, degradation like this goes unnoticed—and unaddressed. Local models ensure that Thai is not just supported, but actively optimized and continuously improved.
More importantly, Kasima drew from over a decade of experience in Silicon Valley to highlight a key reason behind its success. It wasn’t just about cutting-edge technology—it was the ongoing dialogue between developers, users, and researchers that drove rapid iteration and innovation. In that environment, feedback loops are short, ideas are freely exchanged, and communities play an active role in shaping the tools they use.
We don’t want Thailand to be just silent consumers—those who quietly accept whatever tool they're given. We want to build an ecosystem of innovators—those who build, co-create, and actively join the dialogue.
Without that kind of engagement, Kasima warned, users are stuck with tools they can’t improve. That’s why Typhoon is designed to be community-driven to ensure that Thai voices are not only heard, but have a hand in shaping the future of AI that truly understands and serves our local context.
3. Efficiency, Access, and Economic Inclusion: A Dual Perspective from TDRI
Representing both the user’s perspective and a policy research lens, Narin Tananitaporn from TDRI shared how his team successfully deployed Typhoon in a real-world research project demonstrating both its effectiveness and cost efficiency.
The project focused on analyzing Thailand’s job market, where the data came in the form of unstructured job postings. This type of task plays to the strengths of large language models (LLMs), particularly for classification and Named Entity Recognition (NER), as they handle messy, out-of-sample data well. With over 200,000 job posts collected per quarter, the solution also needed to be fast and scalable.
When comparing model options, the team narrowed it down to Typhoon and a foreign open-source LLM as both were significantly more affordable than proprietary alternatives. To evaluate performance, TDRI created a labeled test set then benchmarked both models. The results showed that while both performed similarly overall, Typhoon significantly outperformed in classifying STEM occupations which is a key area for economic development. This makes Typhoon the preferred choice.
Beyond the technical win, Narin highlighted a deeper policy implication: Open Thai language models are essential to building an inclusive AI economy. They give startups, SMEs, and research institutions access to powerful tools that would otherwise be out of reach—lowering the barrier to entry and enabling more players to participate, innovate, and grow.
4. Beyond Just Using AI: Building Knowledge for Long-Term Advantage
Dr. Sarana Nutanong from VISTEC outlined three essential components for sustainable AI advancement: new utilization, new knowledge, and new resources and toolsets. In his talk, he used the following diagram to highlight a crucial gap in Thailand’s AI landscape.
One common path in Thailand is a direct leap from undergraduate talent to industry. While this approach delivers quick wins by getting applications built, it relies heavily on existing (often foreign) tools, without creating original intellectual property or shared resources. Many organizations focus on this “new utilization” stage, building use cases and applications, but few invest in generating new knowledge or developing reusable tools. Without these foundations, long-term progress and competitiveness are limited.
Another path, which Dr. Sarana advocates, develops talent through a research-driven graduate school stage combined with close collaboration with industry. This approach generates knowledge, reusable resources, and toolsets that feed back into industry, delivering not only practical expertise but also assets that can be continuously refined. By pairing utilization with this kind of foundational creation, Thailand can strengthen its entire AI ecosystem and position itself as an active contributor to global innovation rather than a passive consumer of foreign technology.
As Dr. Sarana put it,
Thailand needs R&D team from the industry like Typhoon, one that connects academic research with real industry applications, so innovation flows in both directions.
Conclusion: Building Thailand’s AI Future
The panel clearly highlighted one message: Thailand shouldn’t settle for being a passive user of global AI tools.
With initiatives like Typhoon, local enterprises, developers, and researchers have the opportunity to:
-
Reduce cost and dependency on proprietary models
-
Customize solutions to fit local needs
-
Actively participate in shaping the tools they use
-
Create new knowledge and innovations for the Thai AI ecosystem
The time is now to invest in models and ecosystems that reflect who we are and where we want to go.
