Skip to main content
Back to News
ELYZA releases two commercially usable Japanese models based on LLM-jp-4
Technology
2 min read
JP

ELYZA releases two commercially usable Japanese models based on LLM-jp-4

The AMW Read

The release incrementally updates ELYZA's foundation-model offering with commercially usable Japanese variants, but the reported benchmark gains support a localized impact rather than a frontier shift.
NoveltySignificance
Foundation Models · Player Map
ELYZA
ELYZA

Foundation Models / LLMs

View Company Profile

ELYZA releases two commercially usable Japanese models based on LLM-jp-4

ELYZA, a Tokyo-based AI developer in the KDDI group, released ELYZA-Thinking-1.0-llm-jp-4-33b and ELYZA-Thinking-1.0-llm-jp-4-32b-a3b on October 2. Both are available free on Hugging Face under Apache 2.0, which permits commercial use. The models build on the National Institute of Informatics' open LLM-jp-4 series, with intermediate and post-training focused primarily on Japanese-language performance. ELYZA says they outperform LLM-jp-4.1, announced by NII on September 28, on some benchmarks; the report does not establish a comprehensive performance lead.

The release adds to the Japanese foundation-model ecosystem by pairing a publicly developed base model with company-led specialization and a commercially usable license. Its market relevance lies in that division of work: ELYZA is improving an existing model foundation rather than presenting these releases as newly pretrained models. For builders evaluating Japanese-language systems, the combination makes adaptation and licensing terms concrete considerations alongside benchmark scores. The reported gains remain specific to some benchmarks and should not be read as evidence of superiority across enterprise workloads.

Builders can evaluate the two releases on their own Japanese-language tasks before committing to deployment. Investors should also distinguish the released models from ELYZA's broader research ambition. The company announced ELYZA RSI Research the same day to pursue recursive self-improvement, in which AI helps develop AI through repeated improvement cycles. ELYZA describes the current work as preliminary component research and explicitly says it has not achieved recursive self-improvement itself.

#ELYZA #JapaneseAI #FoundationModels #OpenModels #Apache2

#ELYZA#LLM-jp-4#Japanese language models#Apache 2.0#related:KDDI#related:National Institute of Informatics

How This Connects

Based on Foundation Models · Player Map

  1. 9h agoDeepSeek reportedly nears RMB 80 billion funding round with Tencent and CATLDeepSeek
  2. 1d agoDeepSeek reportedly nears $12 billion round as investor demand lifts its targetDeepSeek
  3. 3d agoAnthropic infrastructure financing reportedly reaches $60B with Broadcom supportAnthropic
  4. 4d agoAnthropic reportedly files confidentially for a potential October 2026 IPOAnthropic
  5. 6d agoELYZA releases two commercially usable Japanese models based on LLM-jp-4 · THIS ARTICLE
  6. 6d agoAnthropic reportedly secures up to $42B in Broadcom financing for AI infrastructureAnthropic

Related News

More news from ELYZA

Stay updated with the latest news and announcements from ELYZA.

View all ELYZA news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard