Ricoh develops multimodal reasoning LLM to make corporate tacit knowledge AI-ready

Ricoh completed a multimodal reasoning LLM base model named Qwen3-VL-Ricoh-32B-20260227 designed to read documents with charts and tables.

Ricoh released Qwen3-VL-Ricoh-8B-20260227 for free and plans to publish a reasoning-performance benchmark tool.

The model is based on Alibaba Cloud's Qwen 3.0-VL, developed under GENIAC Phase 3, and supports multi-step reasoning and Japanese-language thought processes.

Ricoh reduced parameter size for on-premises deployment and supports fine-tuning and private model offerings.