arc_agi_2@1.0

Reasoning
Multimodal

ARC-AGI-2: A benchmark measuring abstract reasoning through visual grid puzzles requiring rule inference and generalization.

← Back to Registry

Run this task

CLI:

inspect eval inspect_harbor/arc_agi_2_1_0 --model openai/gpt-5

Python:

from inspect_ai import eval
from inspect_harbor import arc_agi_2_1_0

eval(arc_agi_2_1_0(), model="openai/gpt-5")

Dataset information

Harbor registry arc_agi_2@1.0
Inspect task arc_agi_2_1_0
Version 1.0
Samples 167
Paper arxiv

See Task Parameters for the parameter set shared across all Harbor tasks.