The Year in AI Papers
The year/Independent research

Paper 2605.24830

Macaron-A2UI: A Model for Generative UI in Personal Agents

Fancy Kong, Congjie Zheng, Murphy Zhuang, Rio Yang, Sueky Zhang, Hao Fu, Gene Jin, Song Cao, Kaijie Chen, Andrew Chen, Pony Ma

Published
May 2026
Research lab
Independent
Citations
1
GitHub
Not linked

01 In brief

Summary

The paper introduces Macaron-A2UI, a model for Generative UI in personal agents, which dynamically synthesizes executable UI actions alongside natural language to improve interaction efficiency.

The authors build a large-scale corpus of over 14,000 samples from four dialogue sources (MultiWOZ, SGD, ESConv, AnnoMI) using a hybrid rule-and-LLM pipeline with deterministic validation, achieving a 99.2% renderability rate.

They also introduce A2UI-Bench, a benchmark with 300 tasks across atomic, depth, and width structures, evaluated via three language-side levels (protocol correctness, task construction, user experience) and three visual-side metrics.

Models of 30B, 235B, and 754B parameters are trained using LoRA-based supervised fine-tuning followed by GRPO reinforcement learning.

The best model, Macaron-A2UI-Venti (754B), achieves an overall score of 75.6 on A2UI-Bench without schema hints, surpassing the strongest full-schema frontier baseline (GPT-5.4 at 74.1).

Results show that SFT and RL significantly improve performance, with RL enhancing protocol correctness first and higher-level interaction quality gradually.

The work demonstrates that Generative UI competence can be internalized without heavy schema prompting at inference time.

02 From the paper

Abstract

As personal agents evolve to handle complex, user-centric tasks, static plain-text chat is rapidly becoming a bottleneck. Generative UI emerges as the necessary new interface layer, dynamically synthesizing the right controls, options, and state from the interaction context in real time. We present Macaron-A2UI, a model for Generative UI in personal agents. Our goal is to move beyond text-only interaction by enabling agents to generate natural language together with lightweight, executable UI actions for information collection, preference refinement, confirmation, and multi-goal organization. We build a large-scale Generative UI corpus from heterogeneous dialogue sources, introduce A2UI-Bench for controlled evaluation, and train 30B, 235B and 754B models with parameter-efficient LoRA-based supervised fine-tuning followed by reward-driven reinforcement learning. The best Macaron-A2UI model reaches 75.6 overall on A2UI-Bench without explicit schema hints, surpassing the strongest full-schema frontier baseline. We release the models, benchmark, and evaluation protocol to support future work on Generative UI for personal agents.