AI Paper Atlas

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

InteractScience: Programmatic and Visually-Grounded Evaluation of Interactive Scientific Demonstration Code Generation

Qiaosheng Chen, Yang Liu, Lei Li, Kai Chen, Qipeng Guo, Gong Cheng, Fei Yuan

ICML 2026 · Large language models › Agents, code and math · approach: LLMs for scientific discovery

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
add reference snapshots as input (ablation)GPT-5 · InteractScience PFTOverall pass rate (%): 39.47 -> 42.53Improved
add reference snapshots as input (ablation)GPT-5 · InteractScience VQTCLIP: 71.95 -> 73.51Improved
add reference snapshots as input (ablation)Qwen2.5-VL-72B-Instruct · InteractSciencePFT and VQTHurt
add reference snapshot to VLM judge input (ablation)VLM-Judge (Igen+L) · Human alignmentSpearman correlation: 0.8224 -> 0.8827Improved
add checklist to VLM judge input (ablation)VLM-Judge (Igen+Iref) · Human alignmentSpearman correlation: 0.3837 -> 0.8827Improved
add checklist to VLM judge with generated image only (ablation)VLM-Judge (Igen) · Human alignmentSpearman correlation: 0.7360 -> 0.8224Improved
add reference snapshot to generated image only judge (ablation)VLM-Judge (Igen) · Human alignmentSpearman correlation: 0.7360 -> 0.3837Hurt
give judge rendered screenshot instead of code only (ablation)VLM-Judge · Human alignmentSpearman correlation: 0.1408 -> 0.7360Improved
swap judge from Gemini-2.5-Pro to GPT-5 / Kimi-K2.5 (ablation)VQT ranking · InteractScienceSpearman rho between rankings: 0.937, 0.919, 0.956No clear effect
scale Qwen3 from 8B to 235BQwen3 · InteractScience PFTOverall pass rate (%): 20.00 -> 33.33Improved
use closed-source instead of open-source modelsLLMs · InteractScience PFTOverall pass rate (%): 41.47 -> 33.87Improved
PFT ability vs visual fidelityper-problem scores · InteractScienceSpearman PFT vs CLIP / VLM-Judge: 0.023 -> 0.312 rangeNo clear effect

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