Web generation improves by learning execution feedback loops
Learning the Loop, Not Just the Page: Execution-Grounded Loop Learning for Web Generation
Computation and Language
Summary
Generating functional web pages is hard because models that create the pages often don't learn well from what happens when the pages run. This paper introduces WebLoop, a method that teaches a model not just to create a page but also to judge and improve its own work by running the page and learning from the results. By training a part called the Critic that does not execute code but learns from complementary signals, the system can better identify and fix bugs. This approach makes web generation more accurate and works well even with different input types and model sizes.
What this means in practice
- •For web developers: Automatically generate web pages with improved accuracy by learning from execution feedback within generation loops.
- •For software engineering teams: Diagnose and refine imperfect code implementations by integrating execution-grounded critics together with generation models.
Authors
Yuxin Meng, Ruixu Zhang, Junjie Wang, Yuhan Suo, Yuhan Sun, Ruining Hu, Yiyao Yu, Yubin Wang, Shouwei Ruan, Bin Wang, Yue Liao, Yuxiang Zhang, Yujiu Yang
Abstract
Functional Web generation is increasingly optimized with executable rewards, yet existing methods largely focus on the quality of the final page and leave the process of diagnosing and repairing imperfect implementations underexplored. We identify a central challenge in this setting: the Generator and Refiner produce executable artifacts with direct environment rewards, whereas the intermediate Critic influences downstream behavior without a directly executable outcome. We introduce WebLoop, an execution-grounded framework that jointly learns generation, critique, and refinement within a shared policy. WebLoop trains an execution-free Critic with complementary signals for requirement-level discriminability and downstream helpfulness, first establishing reliable diagnosis and then introducing consequence-aware credit, while all three roles are jointly optimized with group-relative policy learning. With Qwen3.5-9B, WebLoop reaches 41.5 Overall on WebRise and 38.9% accuracy on WebGen-Bench, improving the base model by 11.3 and 15.4 points, respectively. The gains transfer to first-pass generation, persist at 27B scale, and generalize from text-only training to multimodal inputs. Controlled analyses further show that the improvement cannot be explained by an additional refinement pass alone, highlighting the importance of learning the Critic and the loop itself.