Ruyi2.5 Technical Report

TL;DR

Ruyi2.5 introduces a multimodal model with a shared architecture for scalable deployment, privacy-preserving features, and accelerated reinforcement learning, demonstrating competitive performance on benchmarks and surveillance tasks.

Contribution

It extends the 'Train Once, Deploy Many' paradigm to multimodal models, introduces a privacy-preserving camera system, and proposes BPPO for faster reinforcement learning.

Findings

Ruyi2.5 matches Qwen3-VL on multimodal benchmarks.

Ruyi2.5-Camera outperforms Qwen3-VL on privacy-sensitive tasks.

BPPO accelerates reinforcement learning by 2-3 times.

Abstract

We present Ruyi2.5, a multimodal familial model built on the AI Flow framework. Extending Ruyi2's "Train Once, Deploy Many" paradigm to the multimodal domain, Ruyi2.5 constructs a shared-backbone architecture that co-trains models of varying scales within a single unified pipeline, ensuring semantic consistency across all deployment tiers. Built upon Ruyi2.5, Ruyi2.5-Camera model is developed as a privacy-preserving camera service system, which instantiates Ruyi2.5-Camera into a two-stage recognition pipeline: an edge model applies information-bottleneck-guided irreversible feature mapping to de-identify raw frames at the source, while a cloud model performs deep behavior reasoning. To accelerate reinforcement learning fine-tuning, we further propose Binary Prefix Policy Optimization (BPPO), which reduces sample redundancy via binary response selection and focuses gradient updates on response prefixes, achieving a 2 to 3 times training speedup over GRPO. Experiments show Ruyi2.5 matches Qwen3-VL on the general multimodal benchmarks, while Ruyi2.5-Camera substantially outperforms Qwen3-VL on privacy-constrained surveillance tasks.