ELYZA
Japan's Japanese-first LLM — from U Tokyo's Matsuo Lab, backed by KDDI
Overview
ELYZA, a 2018 University of Tokyo (Matsuo Lab) spinout now 53%-owned by KDDI, builds Japanese-first LLMs — the open Llama-3-ELYZA-JP-8B and enterprise tools (ELYZA Works, App Platform) — sold entirely through direct contract with no public pricing or self-service API.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete.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.
- ELYZA-japanese-Llama models on Hugging Face
- Self-hosted, bring your own compute
- License terms per model card
- No public price list
- Custom Japanese-LLM deployments
- Targets BFSI, telecom, industry and government
- Contact sales via elyza.ai
Use cases
What you can produce with ELYZA
- Draft customer-support email responses in Japanese, cutting operator handling time by up to 60%
- Standardised electronic medical records reformatted from free-text clinical notes
- A deployed internal AI app built without coding, using company data as context, shareable across teams
- A private-cloud LLM endpoint running ELYZA weights on your own infrastructure to support Japan APPI data-residency requirements
- A fine-tuned Japanese domain model (finance, legal, retail) built on the Llama-3-ELYZA-JP-8B base
- Automated structured summaries of Japanese-language call transcripts for CRM integration
- A RAG pipeline over proprietary Japanese documents delivering factual Q&A with source citations
ASEAN Perspective
ELYZA in Southeast Asia
ELYZA is architecturally APAC-relevant but operationally Japan-centric. Japanese enterprises with ASEAN subsidiaries — particularly in finance, telco, and retail — can benefit from ELYZA's integration with KDDI's Japan-hosted GPU infrastructure, which addresses data-sovereignty requirements under Japan's APPI and can underpin intra-group ASEAN workflows routed through a Japanese data centre. ASEAN organisations looking to self-host can download open-weight models (Llama-3-ELYZA-JP-8B, ELYZA-Thinking-1.0-Qwen-32B) directly from Hugging Face for private-cloud deployment, eliminating cross-border data transfer concerns. However, ELYZA has no confirmed ASEAN office, no Bahasa/Thai/Vietnamese fine-tuning, and no regional reseller channel as of mid-2026, so direct procurement from Southeast Asia remains procedurally challenging.
ELYZA has the cleanest academic-to-enterprise pipeline among Japan's domestic LLM shops: University of Tokyo Matsuo Lab research, a 2024 KDDI majority buy-in (53.4% combined stake), and production deployments that aren't demoware — Sumitomo Mitsui Card's contact centre, a Cabinet Office medical LLM, and, as of March 2026, a spot in the Digital Agency's government AI trial alongside KDDI. The open Llama-3-ELYZA-JP-8B holds real community traction on Hugging Face and reportedly matches GPT-3.5-class fluency in Japanese. The catch is that ELYZA is built for Japan and nobody else. No public pricing, no self-service sign-up, no documented API — every relationship starts with a Japanese-speaking sales conversation. Technical teams willing to self-host the open weights get an escape hatch; everyone else, especially outside Japan, will find the front door hard to locate.
What people say
Seven thousand monthly downloads and 148 likes on Hugging Face — modest by global standards, but it's the clearest independent signal ELYZA has, since there's no G2 or Capterra presence and no public review base to speak of. The company, a 2018 spinout from Professor Yutaka Matsuo's University of Tokyo lab, builds Japanese-first LLMs and became a KDDI-controlled subsidiary in April 2024 (KDDI plus KDDI Digital Divergence hold 53.4% combined) — a deal read locally as a 'swing-by IPO' play, parking the startup inside a telco before an eventual public listing.
The open-weight Llama-3-ELYZA-JP-8B is the accessible part of the business: benchmarked around GPT-3.5-class Japanese fluency, downloadable and self-hostable with no sales call required. Everything else runs through direct enterprise contract — ELYZA Works (a no-code app builder relaunched with a KDDI-branded ad campaign in May 2026), the managed App Platform, and named deployments at Sumitomo Mitsui Card and, as of March 2026, a Digital Agency government-AI pilot alongside KDDI.
Practitioners' main complaints are consistent: no published pricing, no self-service API, and a closed demo that makes early evaluation slow. That's by design — ELYZA sells to large Japanese enterprises and government bodies, not developers experimenting on a weekend, and the KDDI backing gives it the compute and distribution to make that model work domestically even as it stays largely invisible outside Japan.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- ELYZA Pencil, the company's first public product, attracted 110,000 users in just 11 days after launch in 2022 — before generative AI became a global phenomenon.
- ELYZA's founding team comes from the Matsuo Lab at the University of Tokyo, one of Japan's most prolific AI spinout factories — the same lab that helped shape Japan's national AI strategy.
- KDDI paid for a 53.4% stake in ELYZA in April 2024, making a Japanese telco the controlling shareholder of the country's most prominent domestic LLM lab.
- The free public demo of ELYZA LLM for JP — built on Llama 2 and then Llama 3 — ran for over two years before being quietly closed in May 2026 as the company pivoted fully to enterprise contracts.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including ELYZA's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with ELYZA 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 ELYZA directly →
Spotted something out of date? Suggest an update →
More in LLMs & Chat
