The Year in AI Papers
The year/Independent research

Paper 2509.18154

MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe

Tianyu Yu, Zefan Wang, Chongyi Wang, Fuwei Huang, Wenshuo Ma, Zhihui He, Tianchi Cai, Weize Chen, Yuxiang Huang, Yuanqian Zhao, Bokai Xu, Junbo Cui, et al.

Published
Sep 2025
Research lab
Independent
Citations
124
GitHub
26K stars

01 In brief

Summary

MiniCPM-V 4.5 is an 8B-parameter multimodal large language model (MLLM) designed for high efficiency and strong performance.

It introduces three core improvements: a unified 3D-Resampler architecture that compresses images and videos into compact token sequences (up to 16x for images and an additional 6x for videos), a unified learning paradigm for document knowledge and OCR that uses dynamic visual corruption to train the model directly on document images without external parsers, and a hybrid reinforcement learning strategy that jointly optimizes short and long reasoning modes.

In OpenCompass evaluation, MiniCPM-V 4.5 achieves a score of 77.0, surpassing GPT-4o-latest and Qwen2.5-VL 72B.

On VideoMME, it achieves state-of-the-art performance among models under 30B, using only 46.7% GPU memory and 8.7% inference time of Qwen2.5-VL 7B.

The model also excels in OCR, document parsing, and hallucination reduction.

Ablations confirm the effectiveness of the hybrid RL strategy, probability-based rewards, unified document learning, and the 3D-Resampler.

02 From the paper

Abstract

Multimodal Large Language Models (MLLMs) are undergoing rapid progress and represent the frontier of AI development. However, their training and inference efficiency have emerged as a core bottleneck in making MLLMs more accessible and scalable. To address the challenges, we present MiniCPM-V 4.5, an 8B parameter model designed for high efficiency and strong performance. We introduce three core improvements in model architecture, data strategy and training method: a unified 3D-Resampler model architecture for highly compact encoding over images and videos, a unified learning paradigm for document knowledge and text recognition without heavy data engineering, and a hybrid reinforcement learning strategy for proficiency in both short and long reasoning modes. Comprehensive experimental results in OpenCompass evaluation show that MiniCPM-V 4.5 surpasses widely used proprietary models such as GPT-4o-latest, and significantly larger open-source models such as Qwen2.5-VL 72B. Notably, the strong performance is achieved with remarkable efficiency. For example, on the widely adopted VideoMME benchmark, MiniCPM-V 4.5 achieves state-of-the-art performance among models under 30B size, using just 46.7\% GPU memory cost and 8.7\% inference time of Qwen2.5-VL 7B.