UIGEN-T3-Dataset-Extended-Reasoning

Summary

A 7,960-row dataset of UI generation tasks with extended pre- and post-reasoning traces, used to train UIGEN-T3, a hybrid-reasoning UI generation model built on the Qwen3 architecture. Each row contains a question, pre-reasoning, post-reasoning, and an HTML/CSS answer implementing the requested UI component or page (e.g., multilingual chatbot interfaces, heatmaps, onboarding flows, live messaging UIs).

Keywords

ui-generation html css code-generation reasoning synthetic web-ui hybrid-reasoning

Topics

Code / UI Generation

Research notes

• Part of the UIGEN series. UIGEN-T3 supports /think and /no_think modes and was evaluated with UIGenEval, a first-of-its-kind UI generation benchmark.