AI Paper Atlas

Research map / Large language models / Agents, code and math

Bringing Code ALIVE: Optimizing Interactive Frontend Mini-Games via Automated Play and Reinforcement Learning at Scale

Jiajun Zhang, Yuheng Jing, Zeyu Cui, Hao Zheng, Wentao Chen, Kaixin Li, Jiaxi Yang, Tianbao Xie, Zeyao Ma, Tianyi Bai, KaShun SHUM, Lei Zhang and 7 more

ICML 2026 · Large language models › Agents, code and math · approach: Software engineering agents

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
curate SFT data with ALIVE-Score (best of five Gemini-3-Pro candidates) and fine-tuneQwen3-Next · ALIVE-EvalALIVE-Score: 25.4 -> 29.1 (+3.7)Improved
curate SFT data with ALIVE-Score and fine-tuneQwen3-Next · ArtifactsBench-Gamescore: 49.2 -> 58.4 (+9.2)Improved
add GRPO RL with ALIVE reward after SFTALIVE-Coder-SFT · ALIVE-EvalALIVE-Score: 29.1 -> 40.5Improved
add GRPO RL with ALIVE reward after SFTALIVE-Coder-SFT · ArtifactsBench-Gamescore: 58.4 -> 61.2Improved
use ALIVE as RL reward instead of Hybrid Judge (code+visual static judge) (ablation)RL on ALIVE-Coder-SFT · ALIVE-Eval test scoretest score: Hybrid test score collapses; response length 7k -> over 14k tokensImproved
use ALIVE as RL reward instead of Visual Judge (screenshots) (ablation)RL on ALIVE-Coder-SFT · ALIVE-Eval test scoretest scoreImproved
ALIVE one-shot planning instead of GUI agent serial loopevaluation framework · evaluation costtoken consumption / latency: approximately 12× lower token consumption and 22× lower latencyImproved
ALIVE instead of GUI agent for RL reward computationRL reward pipeline · 4,500 prompts x 16 samplescost ($): 4,800 -> 396Improved
allow up to three replanning attempts versus one (ablation)ALIVE · evaluation accuracyaccuracy parity with GUI agent steps: 9-10 steps (Replan=1) -> 11-12 steps (Replan=3)Improved
add replanning mechanism in SFT data curation (ablation)ALIVE · SFT dataset selectionfraction of samples rectified: 6% of the total dataset rectifiedImproved
use ALIVE planning instead of heuristic random monkey policies (ablation)ALIVE · fully functional gamesmechanism exposureImproved
use ALIVE instead of code/visual/hybrid judges for human alignment (ablation)evaluation judge · 500 mini-game queries preliminary studyRank Agreement / Score Precision: baselines Rank Agreement <= 0.5; ALIVE > 0.70, Score Precision > 0.85Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy