GoTo releases Indonesian AI model for local languages

For investors, the significance is less about the model’s size than what it says about the direction of Indonesia’s digital economy. GoTo is positioning itself as a platform for domestic AI development rather than just a consumer of imported models, a strategy that could deepen engagement across its e-commerce, ride-hailing and fintech ecosystem if local developers adopt the technology.

Chief Executive Hans Patuwo said Indonesia needs to develop its own AI to solve problems specific to the country, including everyday use of regional languages. GoTo said the model was trained on Indonesian text from media partners and public-sector websites, and it worked with Universitas Sumatera Utara to gather Batak data reflecting current usage.

The launch also underscores the push for more accessible AI deployment in markets where data-center capacity remains limited. GoTo said the model can run on university, startup and public-institution servers without requiring a large-scale data center, though it still needs roughly 54 GB to 60 GB of GPU memory in FP16 or BF16 precision, with Nvidia’s A100 80 GB listed as a recommended setup.

The model’s 128,000-token context window gives it enough room to process long documents or extended conversations, a feature that could appeal to public-sector and enterprise users in legal, education and customer-service applications. But like other generative AI systems, it can still hallucinate and will need testing before it is used in production.

GoTo’s open-source release follows an AI development effort that began in 2024 and involved interns from seven Indonesian universities in data collection, training and testing. The broader implication is that regional-language AI could become a competitive layer in Indonesia’s software market, where localization, cost and trust may matter as much as raw model performance.