Ricoh unveils 35B AI model for Japanese enterprise agents
TL;DR
Ricoh announced RICOH LMM_35B-A3B on October 15, a Japanese-focused multimodal model with approximately 35 billion total parameters and 3 billion active parameters. Ricoh plans to add it to its on-premises LLM starter kit beginning in November and reports improved Japanese agent-task performance over its base model.
About this summary
This article aggregates reporting from 2 news sources. The TL;DR is AI-generated from original reporting. Race to AGI's analysis provides editorial context on implications for AGI development.
Read next
Race to AGI Analysis
Analyst assessment: Ricoh's announcement matters as a test of whether useful agent capabilities can be packaged for local enterprise deployment. The company combines a sparse model architecture with Japanese-focused training and a planned on-premises product. Its reported benchmark improvements concern document understanding and tool-based customer-service tasks, rather than evidence of general intelligence. That distinction matters: practical task completion can improve without demonstrating a broad advance toward AGI. ([jp.ricoh.com](https://jp.ricoh.com/release/2026/1015_1))
The competitive implication is a possible route for enterprise specialists to build value on top of an existing model rather than compete directly in frontier pretraining. Ricoh's use of an Alibaba Cloud base model makes adaptation, deployment and customer access central to that strategy. For Race to AGI readers, the useful question is whether these gains survive unfamiliar documents, changing policies and longer workflows. Independent testing should examine failure recovery and operating costs alongside headline accuracy. If those results hold, this approach could make capable agents accessible to more organizations. The announcement alone does not establish that outcome or justify moving an AGI arrival estimate. ([jp.ricoh.com](https://jp.ricoh.com/release/2026/1015_1))
