ARC-AGI-2
by ARC Prize
What is ARC-AGI-2? ARC-AGI-2 is a reasoning & knowledge eval made by ARC Prize. Grid puzzles where the solver sees a few input-output examples, infers the hidden rule and must draw the exact output grid, with two tries per test input. The repo has 1,000 public training tasks and 120 public evaluation tasks; people in the creators' test sample averaged 66% on the evaluation set, and two private sets are kept back to limit leakage.
Editor note
Grid puzzles where the solver sees a few input-output examples, infers the hidden rule and must draw the exact output grid, with two tries per test input. The repo has 1,000 public training tasks and 120 public evaluation tasks; people in the creators' test sample averaged 66% on the evaluation set, and two private sets are kept back to limit leakage.
- What it tests
- Abstract reasoning and adapting to new problems from very few examples.
- Original source
- github.com/arcprize/ARC-AGI-2
- Last updated
- Repo last commit 2025-05-15
- License
- Apache-2.0
Reviews
No reviews yet.
Reviews from practitioners who sign in with GitHub open soon.
Related reasoning & knowledge evals
- GPQA — Expert-level science reasoning.
- Humanity's Last Exam — Frontier academic knowledge and reasoning.
- LiveBench — Hard, contamination-limited tasks with verifiable ground-truth answers (no LLM judge).
- MMLU — World knowledge and problem solving across many school and professional subjects.