AI Paper Atlas

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

A²RBench: An Automatic Paradigm for Formally Verifiable Abstract Reasoning Benchmark Generation

Qingchuan Ma, Yuexiao Ma, Yongkang Xie, Tianyu Xie, Xiawu Zheng, Rongrong Ji

ICML 2026 · Large language models › Agents, code and math · approach: Inductive and abstract reasoning evaluation

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
remap symbol tokens in symbolic tasks (ablation)GPT-5 · A2RBench symbolic tasksaccuracy: 41.3 -> 23.6Hurt
remap symbol tokens in symbolic tasks (ablation)Claude-Sonnet-4.5 · A2RBench symbolic tasksaccuracy: 34.2 -> 32.5No clear effect
augment seed tasks with expanded input variations (V1-9)solver LLMs · A2RBenchaccuracy: Gemini3-Pro 39.8 -> 41.0No clear effect
use 3D tasks instead of 2D taskssolver LLMs · A2RBenchaccuracyImproved
use highly structured, higher-complexity inputs (variation V4)solver LLMs · A2RBenchaccuracy: 54.8%, nearly triple the baselineImproved
reuse validated seed code to expand tasks instead of generating new seedsA2RBench pipeline · generation costcost per task (USD): $0.19 -> $0.005Improved

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