Sahabat-AI
Open-weight Indonesian LLM built for data sovereignty, not benchmarks
Overview
GoTo and Indosat Ooredoo Hutchison's open-source LLM for Bahasa Indonesia, now a 70B-parameter Llama 3.1 derivative covering five local languages, with weights on Hugging Face and infrastructure kept entirely inside Indonesia.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Jul 2026 and may be out of date or incomplete.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
- 70B-parameter open-weight model (Llama 3.1 continual pretrain)
- Free multilingual chat app via sahabat-ai.com and GoPay
- Covers Bahasa Indonesia, Javanese, Sundanese, Balinese, Batak Toba
- 128k context window (v2 release)
- Weights published on Hugging Face under Llama 3.1 Community License
- Data and GPU infrastructure hosted within Indonesia
- Published IndoMMLU, SEA-HELM, SEA-IFEval, SEA-MTBench benchmark scores
ASEAN Perspective
Sahabat-AI in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Sahabat-AI is Indonesia's flagship sovereign-AI project, built by GoTo and Indosat Ooredoo Hutchison with AI Singapore and four Indonesian universities. The pitch: keep the data, GPUs and weights inside the country, and cover languages global labs skip — Javanese, Sundanese, Balinese, Batak Toba — alongside Bahasa Indonesia. The v2 release, a 70-billion-parameter Llama 3.1 continual-pretrain with a 128k context window, is a real jump from the original 8B/9B fine-tunes, and every checkpoint ships openly on Hugging Face under the Llama 3.1 Community License.
It's not a frontier model and was never built to be one — gains on IndoMMLU and SEA-HELM are incremental, and there's no polished developer API beyond the free chat app and raw weights. For teams that need genuine Bahasa Indonesia fluency, regional-dialect coverage or on-shore deployment for compliance reasons, it's the most credible free option in the country; for general-purpose chat against GPT- or Claude-class models it isn't the target use case.
What people say
Coverage of Sahabat-AI skews toward trade press rather than end-user reviews — there's no G2 or Capterra category for regional sovereign LLMs — but the reaction from Indonesian tech press and hands-on bloggers gives a decent read on how it lands.
Fortune's Asia desk and Light Reading covered the mid-2025 v2 launch as a 'sovereign AI' story more than a product story: the news is that Indonesia now has a 70B-parameter model trained and served on domestic infrastructure, not that it beats ChatGPT. Telecom Review Asia and The Fast Mode framed their coverage the same way, around GPU sovereignty and the five-language scope (Bahasa Indonesia, Javanese, Sundanese, Balinese, Batak Toba) rather than raw capability.
Hands-on write-ups are thinner. A widely shared Indonesian-language Medium post ('Mari Mencoba AI Lokal: Sahabat AI') tested the chat app directly and found it usable for everyday Bahasa Indonesia conversation and regional-dialect questions mainstream chatbots handle poorly, but noticeably behind GPT-4-class models on reasoning-heavy prompts — consistent with an 8-9B continual-pretrain base ahead of the 70B v2 release.
On Hugging Face, the GoToCompany org's model cards show real developer engagement — download counts and discussion threads on the Llama3-8B and Gemma2-9B checkpoints — rather than passive listings, and the project publishes its IndoMMLU, SEA-HELM, SEA-IFEval and SEA-MTBench scores openly instead of only marketing wins. That transparency counts for something relative to closed regional models.
There's no meaningful volume of consumer app-store reviews for the GoPay-embedded chat experience, and no enterprise review-site presence (nothing on G2, TrustRadius or Capterra), which tracks for a government-adjacent open-weights research release rather than a commercial SaaS product. The honest read: well-regarded within Indonesia's AI research and policy community as a sovereignty milestone, lightly used as a daily-driver chatbot, and not yet benchmarked seriously against frontier models by independent third parties.
One detail that recurs across the trade coverage: the training-data partners include University of Indonesia, Gadjah Mada University, Bandung Institute of Technology and Bogor Institute of Agriculture, plus media groups Kompas and Republika supplying Indonesian-language corpora — a domestic-sourcing story that's part of the project's pitch as much as the model weights themselves.
Summary of public user & expert reviews, compiled by RECATOOLS.
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
Cloud & Infra
The Cloud Market Just Crossed a Half-Trillion-Dollar Run Rate — and AI Is Why It's Still A...
Privacy & Data
ASEAN's DEFA: The Region's Bet to Harmonise Cross-Border Data Rules Heads Toward a 2026 Si...
Cloud & Infra
ASEAN's Data-Centre Boom Is Hitting a Wall of Power and Water — and Johor Is Already Turni...
Alternatives to Sahabat-AI
More in LLMs & Chat
