UI-Venus-1.5 Technical Report

TL;DR

UI-Venus-1.5 presents a unified end-to-end GUI agent trained with a four-stage pipeline—Mid-Training, Offline-RL, Online-RL, and Model Merge—achieving state-of-the-art results on multiple GUI grounding and navigation benchmarks and robust real-world navigation in Chinese apps. It introduces a 10B-token Mid-Training corpus across 30+ GUI datasets, a scalable online RL framework via Device-as-a-Service (DaaS), and a model merging strategy to fuse domain-specific grounding, web, and mobile capabilities into a single checkpoint. Empirical results demonstrate strong performance across ScreenSpot-Pro, VenusBench-GD, AndroidWorld, WebVoyager, and other benchmarks, with notable gains from model scaling and an ablation study highlighting the value of each training stage. Collectively, the work delivers a practical, end-to-end GUI assistant capable of autonomously executing complex tasks across diverse web and mobile environments, including 40+ Chinese apps, while maintaining deployment efficiency.

Abstract

GUI agents have emerged as a powerful paradigm for automating interactions in digital environments, yet achieving both broad generality and consistently strong task performance remains challenging.In this report, we present UI-Venus-1.5, a unified, end-to-end GUI Agent designed for robust real-world applications.The proposed model family comprises two dense variants (2B and 8B) and one mixture-of-experts variant (30B-A3B) to meet various downstream application scenarios.Compared to our previous version, UI-Venus-1.5 introduces three key technical advances: (1) a comprehensive Mid-Training stage leveraging 10 billion tokens across 30+ datasets to establish foundational GUI semantics; (2) Online Reinforcement Learning with full-trajectory rollouts, aligning training objectives with long-horizon, dynamic navigation in large-scale environments; and (3) a single unified GUI Agent constructed via Model Merging, which synthesizes domain-specific models (grounding, web, and mobile) into one cohesive checkpoint. Extensive evaluations demonstrate that UI-Venus-1.5 establishes new state-of-the-art performance on benchmarks such as ScreenSpot-Pro (69.6%), VenusBench-GD (75.0%), and AndroidWorld (77.6%), significantly outperforming previous strong baselines. In addition, UI-Venus-1.5 demonstrates robust navigation capabilities across a variety of Chinese mobile apps, effectively executing user instructions in real-world scenarios. Code: https://github.com/inclusionAI/UI-Venus; Model: https://huggingface.co/collections/inclusionAI/ui-venus

Figures (8)

• Figure 1: UI-Venus-1.5 achieves SOTA performance across multiple GUI grounding and navigation benchmarks. Note that in the three radar charts of grounding, we have normalized the results of the top-performing model to 100% to more clearly differentiate comparisons among various baselines.
• Figure 2: System Overview of UI-Venus-1.5. It operates as an end-to-end GUI Agent that interprets user instructions, perceives interface states through screenshots, and executes interactive actions (e.g., clicking, typing, scrolling) to accomplish tasks across diverse executable environments.
• Figure 3: The Four-Stage Pipeline of UI-Venus-1.5. Starting from Qwen3-VL Series, the model progresses through a multi-stage curriculum: (1) Mid-Training on large-scale GUI data for domain knowledge injection; (2) Offline-RL for task-specific optimization across grounding, mobile, and web objectives; (3) Online-RL to enhance navigation in complex, real-world settings; and (4) Model Merge, which unifies the specialized models into the final UI-Venus-1.5.
• Figure 4: (a) The inner part represents the functional task categories (e.g., GUI-VQA, Grounding, Perception), while the outer one details the distribution of specific data sources and target platforms (Web, Desktop, Mobile); (b) Iterative data refinement pipeline with teacher scoring, trace rewriting/reconstruction, and manual verification.
• Figure 5: Data generation loop via DaaS environment. By iteratively performing this pipeline, the success rate of total trace generation raises from 17.9% to over 70%.