Rocket - Think it. Type it. Launch it.
Mini toolsYuan 3.0 40B
By YuanLab
Yuan 3.0 Flash is built as a sparse Mixture-of-Experts LLM that supports text and image inputs while remaining cost-efficient at inference. Only a subset of its 40B parameters is active per step, and a RAPO reinforcement learning scheme is used to improve reasoning accuracy while reducing token usage. The model is aimed at enterprise applications like office automation, customer service, and analytics, providing strong general and multimodal performance with controllable latency and cost.
New
Multimodal
Gen 3
Overview
Yuan 3.0 Flash is a 40B MoE multimodal foundation model from YuanLab that activates about 3.7B parameters per token, targeting enterprise reasoning with lower compute per token.
About YuanLab
YuanLab builds Yuan 3.0, an open-source multimodal foundation model for real enterprise work. It understands and generates across text, images, tables, and documents, and is optimized for RAG, long-document analysis, and complex reasoning. Using a compute-efficient MoE architecture, it delivers strong performance with lower inference cost and supports self-hosted, commercial deployment.
Website:
yuanlab.ai
Tools using Yuan 3.0 40B
No tools found for this model yet.
Last updated: February 25, 2026
