Abstract reasoning · %
ARC-AGI-1
Abstract grid puzzles (François Chollet) that test learning a new rule from a few examples.
What it measures
% of tasks solved on the semi-private set, together with cost per task.
Limitations
ARC Prize terms restrict reproduction, so only the top entries are cited here. Cost matters: compare at similar cost per task.
Contamination risk: Low
Semi-private set is not published.
Leaderboard
6 models
| # | Model | Provider | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 | Anthropic | 98.5% | 1 | Semi-private set; $2.11/task; CoT“Claude Fable 5 (Max)” | Official leaderboard | Oct 22, 2026 |
| 2 | GPT-6 Astra | OpenAI | 98.5% | 3 | Semi-private set; $0.35/task; CoT“GPT-6 Astra (XHigh)” | Official leaderboard | Oct 22, 2026 |
| 3 | Claude Opus 5.5 | Anthropic | 98.5% | 5 | Semi-private set; $0.16/task; CoT“Claude Opus 5.5 (High)” | Official leaderboard | Oct 22, 2026 |
| 4 | Gemini 3.8 Flash | 98.5% | 6 | Semi-private set; $0.21/task; CoT“Gemini 3.8 Flash (High)” | Official leaderboard | Oct 22, 2026 | |
| 5 | GPT-6.1 Sol | OpenAI | 98.5% | 7 | Semi-private set; $0.07/task; CoT“GPT-6.1 Sol (XHigh)” | Official leaderboard | Oct 22, 2026 |
| 6 | Gemini 4 Argon | 98.5% | 9 | Semi-private set; $0.16/task; CoT“Gemini 4 Argon (High)” | Official leaderboard | Oct 22, 2026 |
Data sources & licences
| Source | Licence / terms | How we reuse it | Last fetched |
|---|---|---|---|
| ARC Prize leaderboard ARC Prize Foundation, arcprize.org/leaderboard | ARC Prize Terms: no copying/aggregation for commercial purposes without permission | Restricted – only top entries cited top 10 per set only (terms) | Oct 22, 2026 |
FAQ
What does ARC-AGI-1 measure?
Abstract grid puzzles (François Chollet) that test learning a new rule from a few examples. % of tasks solved on the semi-private set, together with cost per task.
How reliable is ARC-AGI-1?
ARC Prize terms restrict reproduction, so only the top entries are cited here. Cost matters: compare at similar cost per task. Contamination risk: Low – Semi-private set is not published.