Sahabat-AI
Open-weight Indonesian LLM, now also Indosat's consumer AI app
Overview
Sahabat-AI is an open-weight language-model family for Bahasa Indonesia and regional languages, co-initiated by GoTo and Indosat Ooredoo Hutchison and developed with AI Singapore. Its v2 model is a 70-billion-parameter Llama 3.1 derivative with weights on Hugging Face. Since February 2026 Indosat also runs a Sahabat-AI app for Android and iOS with paid in-app plans.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 19 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Sahabat-AI
- A 70-billion-parameter Llama 3.1-based instruction model with a 128k context, downloadable from Hugging Face
- Smaller v1 Gemma 2 9B and Llama 3 8B models for lighter hardware
- Text generation in Indonesian, Javanese, Sundanese, Balinese, Batak Toba and English
- Published IndoMMLU and SEA-HELM benchmark results to compare against other models
- On-shore deployment, keeping data and inference inside Indonesia
- A consumer app for Android and iOS that routes between Sahabat and frontier models
Sahabat-AI is the most serious open model for Indonesian and its regional languages: the v2 weights cover Javanese, Sundanese, Balinese and Batak Toba alongside Bahasa Indonesia, publish their benchmark scores, and can be run on infrastructure inside Indonesia for teams with data-sovereignty duties. The cost is hardware: about 140 GB of GPU memory at full precision.
Indosat now leads it, and its energy has gone into a consumer app that mixes Sahabat with frontier models and sells token plans. The promised Nemotron-based next generation had not appeared by October 2026, so the model is ageing. Use it for Indonesian-language work; not as a general frontier model.
What we found
Sahabat-AI ("close friend" in Indonesian) is Indonesia's open-weight language-model project. It was introduced at Indonesia AI Day in November 2024, co-initiated by GoTo and Indosat Ooredoo Hutchison, and the Hugging Face model card credits GoTo and AI Singapore as developers. The first release paired Gemma 2 9B and Llama 3 8B models; v2, published on 30 May 2025, is a 70-billion-parameter instruction model built on Llama 3.1 with a 128k context. It needs roughly 140 GB of GPU memory at full precision, so running it yourself is a data-centre job.
Language coverage is the point: the v2 card lists English, Indonesian, Javanese, Sundanese, Batak Toba and Balinese, with training data from six Indonesian universities and media groups including Kompas Gramedia, Republika, Tempo and Hukumonline. IndoMMLU and SEA-HELM scores are published on the card. Licences differ by generation: the Llama 3.1 Community License for v2, Gemma and Llama 3 terms for v1.
Since then the project has become Indosat's. The sahabat-ai.com site calls it an innovation from Indosat, and in February 2026 Indosat and Komdigi, the digital ministry, launched a Sahabat-AI app for Android and iOS. Its App Store listing describes routing between home-grown Sahabat models and frontier models, and later releases added document creation, AiStorage for Pro and Max plans, companions for bill payments and Tri mobile services, and a single token system. In-app purchases on the Indonesian store run from Rp25,000 for Pro to Rp100,000 for Super, with boosters for images, video and code; billing periods are not stated. The listing showed 4.9 from 14 ratings, read 19 October 2026, which is too few to mean much.
What has not shipped is a new model. At NVIDIA GTC in March 2026 Indosat said the next generation would be built on NVIDIA's Nemotron open models and trained on GPUs inside Indonesia. As of October 2026 the newest public weights are still v2 from May 2025.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Sahabat-AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Sahabat-AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Sahabat-AI directly →
Spotted something out of date? Suggest an update →
Sahabat-AI in the news
Alternatives to Sahabat-AI
More in LLMs & Chat
