AI Paper Atlas

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

VisCoder2: Building Multi-Language Visualization Coding Agents

Yuansheng Ni, Songcheng Cai, Xiangchao Chen, Jiarong Liang, Zhiheng Lyu, Jiaqi Deng, Kai Zou, Ping Nie, Fei Yuan, Xiang Yue, Wenhu Chen

ICLR 2026 · Large language models › Agents, code and math · approach: Agent instruction-following benchmarks

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
fine-tune on VisCode-Multi-679KQwen2.5-Coder-7B · VisPlotBench overallExecution pass rate (%): 51.2 -> 70.9Improved
fine-tune on VisCode-Multi-679KQwen2.5-Coder-32B · VisPlotBench overallExecution pass rate (%): 57.5 -> 73.1Improved
add multi-round self-debug (up to 3 rounds)VisCoder2-32B · VisPlotBench overallExecution pass rate (%): 73.1 -> 82.4Improved
add self-debugGPT-4.1 · VisPlotBench LaTeXExecution pass rate (%): 31.3 -> 66.1Improved
add CoSyn synthetic data (ablation)Qwen2.5-Coder-7B · VisPlotBench LilyPondExecution pass rate (%): 5.5 -> 65.5Improved
train only on The-Stack-V2 subset (ablation)Qwen2.5-Coder-7B · VisPlotBench overallExecution pass rate (%): 51.2 -> 49.0Hurt
train only on StarVector SVG subset (ablation)Qwen2.5-Coder-7B · VisPlotBench overallExecution pass rate (%): 51.2 -> 40.1Hurt
add Code-Feedback multi-turn dialogues (ablation)Qwen2.5-Coder-7B · VisPlotBench overall, self-debugExecution pass rate (%): 59.0 -> 63.1Improved
combine all four data sources (ablation)Qwen2.5-Coder-7B · VisPlotBench overallExecution pass rate (%): 59.2 -> 70.9 (best single source CoSyn)Improved
fine-tune on VisCode-Multi-679KQwen2.5-Coder-32B · VisPlotBench SVGExecution pass rate (%): 93.9 -> 81.5Hurt

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