GoTo Open-Sources a 27-Billion-Parameter LLM Built for Indonesian and Regional Languages
GoTo Gojek Tokopedia has released an open-source large language model with 27 billion parameters designed to understand Indonesian and several regional languages, Bisnis.com reported from Bali Tech Week in Sanur on Monday, October 19. Beyond Indonesian, the model can process Javanese, Sundanese, Balinese and Batak, the company said.
Chief executive Hans Patuwo said Indonesia already uses global AI widely but must also build its own technology for national problems, because global models are often not fully optimised for local data or the regional languages people use every day. Access is free, with academics, startups, businesses and public institutions named as the main audiences, which Patuwo framed as GoTo’s contribution to the country’s technology base.
GoTo’s in-house engineers built the model, continuing open-model research the group began in 2024. Training data came from media partnerships, public-sector websites and dedicated collaborations; for Batak, GoTo worked with Universitas Sumatera Utara to record contemporary usage. At 27 billion parameters the model sits in the mid-sized range, which GoTo says is strong enough for complex language tasks yet efficient enough to run on servers at campuses, startups or government agencies without hyperscale data centres. Students from seven universities from Sumatra to Sulawesi, including Universitas Udayana in Bali, joined data processing, training, speech modelling and testing through an internship programme.
What it means in practice
The release came at the same event where Meutya Hafid said Indonesia must set its own AI pace, and it pushes against the mood captured in Jakarta sidelining homegrown foundation models for a compute-first approach. It widens the field of local models after Telkom University’s EduNusa and Medisiana launch, and contrasts with GoTo’s earlier stance in narrowing its AI bets to conversion and cost.
A mid-sized open model is a practical choice for ministries and universities that cannot rent large GPU clusters, but real adoption will depend on licence terms, benchmark results in each regional language, and whether developers trust the data sourcing. The company ties the effort to the Communication and Digital Affairs Ministry’s Indonesia Digital Vision 2045, which aims to turn the country from a consumer into a creator of AI technology.
What to watch: published benchmarks for Javanese, Sundanese, Balinese and Batak, which government agencies deploy the model first, and whether other Indonesian platforms follow GoTo in releasing model weights.