Work

From Pixels to BFS: High Maze Accuracy Does Not Imply Visual Planning

Year: 2026

Type: conference-paper

Abstract: Figure 1: Top: Mazes from MAZEBENCH at increasing difficulty: 5×5 (A) to 20×20 (X).Bottom: Solve rate vs. total tokens.The newly released GPT-5.5 sets a new Pareto point (92% at 1,096 tokens/solve); Claude 4.6-era models (red) cluster at 2-6% and Claude Opus 4.7 reaches 29-31%, still trailing the frontier on image input.

Source: Workshop on Advances in Language and Vision Research (ALVR)

DOI: 10.18653/v1/2026.alvr-main.13

Author: Alberto Gonzalo Rodriguez Salgado

Language: English

Topic: AI-based Problem Solving and Planning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

OpenAlex ID: W7166810860

FWCI: 8.936

Cites:

Cited by: 1

Related to: 10