ARC-AGI Model Analytics

An interactive archive of the ARC Prize team's evaluation results for large language models on ARC-AGI-1 and ARC-AGI-2, as they were published at huggingface.co/arcprize.

When checked on October 7, 2026, the ARC-AGI-1 and ARC-AGI-2 Hugging Face result datasets both listed June 4, 2026 as their last update. This page explores imported records.

Looking for ARC-AGI-3? See the live ARC Prize 2026 Kaggle leaderboard, the mechanics of every official ARC-AGI-3 game, or what we are doing on ARC-AGI-3.

This archive combines official ARC Prize results imported from Hugging Face with community and local runs. Prompts, reasoning settings, available attempts and dataset coverage can differ; compare matching configurations and coverage. For ARC Prize’s evaluation and scoring code, see its open-source benchmarking harness.

What you can do here

• Pick a model and a dataset (ARC-AGI-1, ARC-AGI-2 and others) and see which puzzles it solved, failed or skipped.
• Compare accuracy, cost per task and reliability across models.
• Open any puzzle to see the grids and every model's answer.

Related: head-to-head model comparison, official scoring, browse every puzzle.