A new benchmark shows OpenAI's gpt-5 and gpt-image-1 lose substantial spatial information when transferring colored grids from image to text and back to image.
Evaluating the Ability of Large Language Models to Reason About Cardinal Directions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LMM Modality Transfer: A Pre-requisite for Autonomous GIS Agents
A new benchmark shows OpenAI's gpt-5 and gpt-image-1 lose substantial spatial information when transferring colored grids from image to text and back to image.