Uniphore/actio-ui-7b-rlvr

VISIONPricing:Input $0.4 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:32kPublished:Dec 5, 2025License:openmdw-1.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Uniphore/actio-ui-7b-rlvr is a 7 billion parameter vision-language model developed by Orby AI (a Uniphore company) with a 32768 token context length, specifically trained for web GUI subtask execution. Built upon Qwen2.5-VL-7B-Instruct and fine-tuned with Reinforcement Learning with Verifiable Rewards (RLVR), it achieves state-of-the-art performance among open-source models on the WARC-Bench GUI subtask benchmark. This model excels at short-horizon web interaction tasks like date selection or content extraction within larger browsing workflows, demonstrating improved long-horizon GUI task and grounding capabilities.

Loading preview...

ActIO-UI: Vision-Language Models for GUI Subtasks

ActIO-UI-7B-RLVR is a 7 billion parameter vision-language model developed by Orby AI, a Uniphore company, and introduced in the paper "WARC-Bench: Web Archive based Benchmark for GUI Subtask Executions". This model is built on the Qwen2.5-VL-7B-Instruct base and has been fine-tuned using Reinforcement Learning with Verifiable Rewards (RLVR) to specialize in web GUI subtask execution. It is designed to handle short-horizon web interaction tasks, such as selecting dates in a picker or scrolling to extract information, within broader web browsing workflows.

Key Capabilities

  • Specialized GUI Subtask Execution: Achieves state-of-the-art (SOTA) results for open-source models on the WARC-Bench GUI subtask benchmark, outperforming its base model.
  • Improved Long-Horizon Task Performance: Exhibits enhanced capabilities for long-horizon GUI tasks compared to its base model, despite not being explicitly trained on such data.
  • Vision-Language Integration: Processes visual inputs (images) alongside natural language instructions to interact with graphical user interfaces.
  • Robustness: Demonstrates improved grounding capabilities for GUI elements.

Good For

  • Automating Web Interactions: Ideal for tasks requiring precise interaction with web elements based on natural language commands.
  • Research in GUI Automation: A strong baseline for further research and development in web agent technologies and GUI task execution.
  • Benchmarking: Provides a high-performing open-source model for evaluating GUI subtask performance on benchmarks like WARC-Bench, WebArena, and Miniwob++.

For optimal performance, input images should be set at 1280x720 pixels. The model is licensed under the Open Model, Data, & Weights License Agreement (OpenMDW) and is intended for research and educational purposes.

Finetunes
1 models