Paper 2508.11737
Ovis2.5 Technical Report
- Published
- Aug 2025
- Research lab
- Independent
- Citations
- 81
- GitHub
- 1.5K stars
01 In brief
Summary
Ovis2.5 is a multimodal large language model from Alibaba Group, succeeding Ovis2.
It introduces native-resolution visual perception using a NaViT vision transformer, which processes images at their original resolutions to preserve fine details and global layout, crucial for complex charts.
The model also adds an optional 'thinking mode' for deep reasoning, trained on data that includes reflection, self-checking, and revision.
Training follows a five-phase curriculum: visual pretraining, multimodal pretraining, instruction tuning, DPO, and GRPO, with infrastructure optimizations like data packing and hybrid parallelism yielding a 3-4x speedup.
Two models are released: Ovis2.5-9B and Ovis2.5-2B.
On the OpenCompass multimodal leaderboard, Ovis2.5-9B scores 78.3, a substantial improvement over Ovis2-8B and state-of-the-art among open-source MLLMs under 40B parameters; Ovis2.5-2B scores 73.9, also SOTA for its size.
The models achieve leading results on STEM benchmarks, visual grounding, video tasks, and complex chart analysis, with Ovis2.5-9B outperforming GPT-4o on OCRBench v2.
02 From the paper
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.