Rakuten AI 3.0
Japan's largest homegrown LLM — government-backed, open-weight, built for Japanese-English business use
Overview
Rakuten AI 3.0 is Rakuten Group's flagship Japanese-optimized large language model, developed as part of Japan's government-backed GENIAC project (METI/NEDO). It's a ~671B-parameter mixture-of-experts model (37B active per token, 128K context) tuned for Japanese and English writing, code generation, and document analysis/extraction, and powers Rakuten's own e-commerce, travel, and fintech products internally.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 9 Jul 2026 and may be out of date or incomplete.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 Rakuten AI 3.0
- Download the full 671B-parameter model weights from Rakuten's official Hugging Face repository under the Apache 2.0 licence and deploy them on your own infrastructure with no usage fees.
- Draft, summarise and polish Japanese business documents with a model specifically tuned for Japanese writing conventions and Japan-specific cultural knowledge.
- Feed contracts, reports or manuals of up to 128K tokens into the context window and extract structured data or answers from across the entire document.
- Generate and explain code in a bilingual Japanese-English development workflow, one of the model's stated strengths.
- Fine-tune or continue pre-training the open weights on proprietary industry data to build a domain-specific Japanese model for finance, retail or healthcare.
- Distill the large MoE model into smaller task-specific models that are cheap enough to run in production, as the open licence explicitly permits.
- Translate and adapt content between Japanese and English for business use, leveraging the model's dual-language optimisation.
ASEAN Perspective
Rakuten AI 3.0 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).
What this is for: running a genuinely capable Japanese-English bilingual LLM for writing, coding, and document work, either self-hosted or as the model behind Rakuten's own consumer products. It's Japan's most prominent domestically-built frontier-adjacent model, backed by the government's GENIAC initiative rather than a purely commercial lab.
Who this is for: developers and researchers who want an open-weight model with strong Japanese-language performance to self-host, and anyone tracking Japan's sovereign-AI progress. It is not a Claude/GPT-tier global frontier model — treat it as a strong regional specialist, not a first choice for English-only or non-Japanese workloads.
Availability: fully open — download the weights from Hugging Face and run it yourself, free, no signup or API key required. There is no first-party hosted API as of mid-2026.
What people say
Rakuten AI 3.0 shipped on March 17, 2026, after a December 2025 unveiling, and its headline facts are unusual for a corporate LLM: it is Japan's largest homegrown model — a 671B-parameter mixture-of-experts design activating about 37B parameters per token with a 128K context window — developed under the government-backed GENIAC programme (METI/NEDO) and released open-weight under Apache 2.0 on Hugging Face. That combination of scale, sovereign backing and a genuinely permissive licence is why it drew attention well beyond Japan.
Reception in the Japanese developer and enterprise community has been broadly positive on substance. Rakuten published benchmark results showing strong performance on Japanese-specific evaluations — cultural knowledge, graduate-level reasoning, competitive mathematics and instruction following — and commentators note the model is battle-tested internally, powering features across Rakuten's own e-commerce, travel and fintech products. Japanese analysts frame it as a shift in the domestic AI race from "who can build the biggest model" to "what is actually usable in Japanese workplaces," and credit the open release for enabling distillation, continued training and industry-specific tuning by third parties.
The practical caveats are real, though. A 671B-parameter MoE is far beyond what most organisations can self-host, so hands-on adoption skews toward well-resourced enterprises, cloud deployments and distilled derivatives rather than hobbyists. As a March 2026 release there is not yet a meaningful body of independent user reviews on platforms like G2, and English-language community feedback remains thin; most public evaluation so far comes from Rakuten's own benchmarks and Japanese tech press rather than neutral third parties.
Rakuten AI 3.0 fits Japanese enterprises and developers who need top-tier Japanese-language performance with full weight access — for on-premise deployment, fine-tuning on proprietary data, or building Japanese-English business applications — and researchers who want a large open MoE to study. Teams needing a plug-and-play consumer chatbot should look at hosted alternatives.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Rakuten AI 3.0's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Rakuten AI 3.0 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 Rakuten AI 3.0 directly →
Spotted something out of date? Suggest an update →
Alternatives to Rakuten AI 3.0
More in LLMs & Chat
