🚀 Introduction
This blog aims to provide valuable insights and highlights from the upcoming ASEAN GenAI Startups report, scheduled for release in July. To gather comprehensive insights, we have analyzed over 500 GenAI startups and conducted interviews and surveys with over 120 GenAI startups across Southeast Asia on 25 different data points (this data is still growing). One of the key questions we asked was which foundational model they are using for their Generative AI solutions and why.
By sharing preliminary findings, we seek to offer a glimpse into the current trends and preferences among GenAI startups in Southeast Asia, particularly regarding the foundational models they utilize.
If any startups would like to participate in the survey and get featured in the final report, the deadline for submissions is 14 June: https://bit.ly/genai-report-1
🧠 Understanding Foundational Models
Foundational models are large, pre-trained AI models that can be fine-tuned for specific tasks. Trained on vast datasets, they offer versatility and power for various applications. Developed by top AI organizations, these models can understand and generate text, images, and other data forms, serving as a strong base for specialized AI systems, like GPT for text and DALL-E for images.
📊 Top Foundational Models used by GenAI Startups in Southeast Asia
The chart below depicts the usage distribution of foundational models among Generative AI startups in the ASEAN region. Note: This data is still work in progress as we are still receiving a huge volume of responses.
📈 Top Foundational Models Identified in the Survey
- OpenAI (GPT, DALL E)
- Hugging Face (open-source models)
- Anthropic (Claude)
- Meta (Llama)
- Stability AI
- Google (Gemini)
- Cohere
- Mistral
- Others: This includes Groq, in-house/custom models.
🤔 Why do GenAI startups in Southeast Asia prefer these Foundational Models?
In our survey, we also asked the startups to name the top foundational model provider they would like to work with. Here are the summary of the results:
Foundational Model |
Strengths |
Reason for Preference |
| OpenAI | Advanced AI capabilities, specific applications in AI-driven cybersecurity, and voicebot solutions with good support for the Thai language. | Preferred for its innovation, accuracy, and performance despite higher costs. |
| Hugging Face | Open-source community, customization and fine-tuning, cost efficiency, and compilation for edge use. | Considered for open-source solutions and cost-saving measures. |
| Claude (Anthropic) | High-quality output, larger context windows, performance in generating quality content, and price-to-performance ratio. | Used primarily for industry solutions and exploring other applications. |
| Mistral | Customization, synthetic data generation, and open-source model benefits. | Used for synthetic data generation and customization needs. |
| Meta (LLaMA) | Customization and fine-tuning, scientific R&D support, and cost efficiency. | Used for customization, fine-tuning, and scientific research support. |
| Stability AI | Leading in foundation models for image and video generation. | Used heavily in production for image and video generation. |
| Google Gemini | Notable for upcoming features like 2M contexts as announced at Google IO 2024. | Considered for its potential new features. |
Other Preferences
- Amazon Bedrock: Preferred for multi-modal models.
- Cohere: Multilingual models aligning with support for different languages.
- SeaLLM (Hugging Face): Advanced LLM for SEA languages.
📝 Quick Takeaways
The GenAI startups ecosystem in Southeast Asia is vibrant and diverse, with startups using various foundational models to innovate and solve complex problems. OpenAI, Hugging Face, and Anthropic are the top providers, offering advanced capabilities and strong community support. These models will be crucial in shaping the future of AI-driven solutions in the region.
🔜 What’s Next?
If any startups would like to participate in the survey and get featured in the final report, the deadline for submissions is 14 June: https://bit.ly/genai-report-1
Stay tuned for the next blog in our series, where we will delve deeper into the preferences in Cloud providers of GenAI startups in Southeast Asia! Look forward to the full ASEAN GenAI Startups report, set for release in July, for comprehensive analysis and insights that will help you stay ahead in the rapidly evolving AI landscape.