Ovis2.5 Technical Report

2025-08-15

AI Summary

To address weak native-resolution perception, limited reasoning capability, and low deployment efficiency of multimodal large language models (MLLMs) on high-resolution vision-intensive tasks—such as complex chart analysis, visual grounding, and video understanding—this paper introduces the Ovis2.5 series. Methodologically: (i) we design a native-resolution ViT backbone to preserve fine-grained image details; (ii) we propose a dual-path reasoning architecture integrating linear inference with reflective inference, enabling self-checking and dynamic revision; and (iii) we employ five-stage curriculum learning, multimodal data packing, hybrid parallel training, and alignment optimization via joint DPO and GRPO. Our key contribution is achieving state-of-the-art performance for compact models under resource constraints: Ovis2.5-9B and Ovis2.5-2B attain average scores of 78.3 and 73.9, respectively, on OpenCompass—ranking first among open-source models of comparable scale—and significantly outperform prior work on STEM reasoning, chart comprehension, and video understanding benchmarks.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

We present Ovis2.5, a successor to Ovis2 designed for native-resolution visual perception and strong multimodal reasoning. Ovis2.5 integrates a native-resolution vision transformer that processes images at their native, variable resolutions, avoiding the degradation from fixed-resolution tiling and preserving both fine detail and global layout -- crucial for visually dense content like complex charts. To strengthen reasoning, we train the model to move beyond linear chain-of-thought and perform reflection -- including self-checking and revision. This advanced capability is exposed as an optional "thinking mode" at inference time, allowing users to trade latency for enhanced accuracy on difficult inputs. The model is trained via a comprehensive five-phase curriculum that progressively builds its skills. The process begins with foundational visual and multimodal pretraining, advances through large-scale instruction tuning, and culminates in alignment and reasoning enhancement using DPO and GRPO. To scale these upgrades efficiently, we employ multimodal data packing and hybrid parallelism, yielding a significant end-to-end speedup. We release two open-source models: Ovis2.5-9B and Ovis2.5-2B. The latter continues the "small model, big performance" philosophy of Ovis2, making it ideal for resource-constrained, on-device scenarios. On the OpenCompass multimodal leaderboard, Ovis2.5-9B averages 78.3, marking a substantial improvement over its predecessor, Ovis2-8B, and achieving state-of-the-art results among open-source MLLMs in the sub-40B parameter range; Ovis2.5-2B scores 73.9, establishing SOTA for its size. Beyond aggregate scores, Ovis2.5 achieves leading results on STEM benchmarks, exhibits strong capabilities on grounding and video tasks, and achieves open-source SOTA at its scale for complex chart analysis.

Problem

• Enhancing native-resolution visual perception for dense content
• Advancing multimodal reasoning beyond linear chain-of-thought
• Optimizing model performance for resource-constrained deployment scenarios

Innovation

• Native-resolution vision transformer for detailed visual perception
• Reflection-based reasoning with self-checking and revision
• Five-phase curriculum training with DPO and GRPO