ELYZA releases free fully domestic AI models ELYZA-Thinking-1.0-llm-jp-4-33b, ELYZA-Thinking-1.0-llm-jp-4-32b-a3b

ELYZA, an AI development company under the KDDI group, has released two new AI models that it describes as 'fully domestic infrastructure.' These models are built on the LLM-jp-4 series developed by the National Institute of Informatics (NII), an open AI model series, and have undergone additional training focused mainly on Japanese language processing. ELYZA claims that on some benchmarks, these models outperform the improved version of LLM-jp-4.1 that NII announced on September 28.

The release is part of ELYZA's broader research efforts. On the same day, the company announced the establishment of ELYZA RSI Research, a research organization dedicated to advancing Recursive Self-Improvement (RSI), where AI is used to develop AI to iteratively improve performance. However, ELYZA notes that at this point, the work is at the preliminary research stage and does not yet achieve RSI itself.

The significance of this release hinges on whether these domestically developed models can deliver competitive performance for Japanese language tasks, as claimed by ELYZA. If the benchmarks hold up under independent evaluation, it could offer Japanese businesses a free, commercially usable alternative to foreign AI models. The establishment of ELYZA RSI Research also signals a longer-term ambition in automating AI development, though concrete results from that effort are likely still some way off.