PT GoTo Gojek Tokopedia Tbk (GOTO), Indonesia's largest digital ecosystem, is further cementing its tech sector leadership with the launch of Sahabat-AI, a pioneering open-source Large Language Model (LLM) for Bahasa Indonesia and the country’s regional languages. The initiative, co-developed with PT Indosat Ooredoo Hutchison Tbk, reflects GOTO's commitment to driving technological innovation throughout Indonesia.
Large Language Models (LLMs) are advanced, deep-learning AI programs trained on massive datasets that can understand and produce natural language, enabling a wide range of applications and services.
Patrick Walujo, GoTo Group CEO said: “As a homegrown tech company, we constantly strive to contribute meaningfully to Indonesia's growth. We are proud to partner with Indosat in co-initiating Sahabat-AI, an open-source Large Language Model (LLM) designed to understand the local context and bridge the gaps left by global models. Sahabat-AI will empower the development of AI-based applications and services uniquely tailored to Indonesia and the region. This is aligned with our mission to build Indonesia’s digital sovereignty, as envisioned by our President, Bapak Prabowo Subianto."
Catherine Hindra Sutjahyo, GoTo Group President of On Demand Services (Gojek), speaking today at the Sahabat-AI inauguration event in Jakarta on Indonesia AI Day, said: "We initiated Sahabat-AI because Indonesia, as the fourth most populous country in the world, with its rich cultural diversity and rapid technological advancement, has a clear need for a localised LLM that understands and reflects our culture."
"What sets Sahabat-AI apart from other global LLMs is its deep localisation, specifically designed for the Indonesian language and the country’s various regional languages, such as Javanese and Sundanese. The model brings an in-depth understanding of local context and cultural relevance, empowering inclusivity and promoting digital literacy. In the future, it will expand to include more regional languages such as Batak and Balinese, further enriching its capability.”
Sahabat-AI has been intentionally developed as an open-source ecosystem, allowing local developers and engineers broad access to create AI-based solutions tailored to various needs, including public services, customer service, data analytics, research and development, education, and business growth. The initiative not only benefits Indonesian society at large but also supports the preservation of the country’s regional languages, reducing reliance on foreign AI models. In its first phase, Sahabat-AI will launch with 8-billion and 9-billion parameter LLMs.
“Although we are still in the early stage of development, we are proud to share that the initial model has demonstrated leading performance in Bahasa Indonesia, Javanese and Sundanese, outperforming other open-source models with a similar number of parameters. To go even further and continue to improve the model’s language and cultural comprehension, we are inviting stakeholders from all sectors to collaborate on the development of this uniquely Indonesian AI ecosystem,” added Catherine.
Sahabat-AI LLM model is now available for free download on Hugging Face1, a platform where the machine learning community collaborates on models, datasets and applications.
As an open-source ecosystem, Sahabat-AI connects research institutions, universities, media, government, and other partners to collaboratively power this Indonesian AI technology. In the first phase of GOTO's collaboration with universities in Indonesia, Gadjah Mada University has begun sending their top students to join the Sahabat-AI project team. These students are contributing to the initiative through knowledge sharing, data cleaning, and validation, while also receiving AI training from GOTO.
Sahabat-AI is already powering Dikte Suara (Dira), an AI technology developed by GOTO for its Financial Technology (Fintech) business unit and Gojek business unit. Dira enhances user experience by enabling easier navigation of GoPay features and faster task completion through voice commands in Indonesian. The Dira feature will soon be accessible in the Gojek apps as well.