# ARC-AGI (ARC Prize)

Puzzles that most people find easy but that require spotting a brand-new pattern rather than recalling facts; the newest version (ARC-AGI-3) drops the AI into small interactive games it has never seen. It is a popular test of whether AI can genuinely learn new skills on the fly.

Category: reasoning. Type: leaderboard. Status: active.
Metric: Share of ARC-AGI-3 game environments solved on the semi-private test set (%) (higher is better).
Source: ARC Prize Foundation — https://arcprize.org/media/data/leaderboard/v3.json, results as of 2026-11-16, retrieved 2026-11-17, license: not stated.

| Rank | System | Organization | ARC-AGI-3 score (%) | Total cost (USD) |
|---|---|---|---|---|
| 1 | GPT-6 Astra - Provider Adapter (High) | OpenAI | 99.9 | 18816.63 |
| 2 | Gemini 4 Carbon - Provider Adapter (Medium) | Google | 99.1 | 4559.50 |
| 3 | GPT-6 Astra - Provider Adapter (Max) | OpenAI | 98.6 | 17331.95 |
| 4 | GPT-6 Astra - Provider Adapter (XHigh) | OpenAI | 98.4 | 18147.19 |
| 5 | GPT-6 Astra - Provider Adapter (Medium) | OpenAI | 98.4 | 19284.78 |
| 6 | GPT-6 Astra - Provider Adapter (Low) | OpenAI | 98.0 | 21297.92 |
| 7 | Claude Fable 5.2 - Provider Adapter (High) | Anthropic | 97.7 | 18624.91 |
| 8 | Gemini 4 Carbon - Provider Adapter (Low) | Google | 97.7 | 4350.87 |
| 9 | Gemini 4 Argon - Provider Adapter (Low) | Google | 97.5 | 4770.15 |
| 10 | Gemini 4 Argon - Provider Adapter (Medium) | Google | 97.3 | 4981.20 |
| 11 | Gemini 4 Argon - Provider Adapter (High) | Google | 97.3 | 4625.84 |
| 12 | Claude Fable 5.2 - Provider Adapter (XHigh) | Anthropic | 97.3 | 18251.68 |
| 13 | Gemini 4 Carbon - Provider Adapter (High) | Google | 96.9 | 4350.03 |
| 14 | Claude Fable 5.2 - Provider Adapter (Medium) | Anthropic | 96.8 | 21662.29 |
| 15 | GPT-6 Astra - Provider Adapter (None) | OpenAI | 96.7 | 23456.66 |
| 16 | GPT-6.1 Sol - Provider Adapter (XHigh) | OpenAI | 96.4 | 4360.21 |
| 17 | GPT-6.1 Sol - Provider Adapter (Max) | OpenAI | 96.2 | 3816.80 |
| 18 | GPT-6.1 Sol - Provider Adapter (High) | OpenAI | 95.0 | 5403.98 |
| 19 | Claude Fable 5.2 - Provider Adapter (Max) | Anthropic | 94.9 | 17875.84 |
| 20 | Claude Fable 5.2 - Provider Adapter (Low) | Anthropic | 92.3 | 27377.33 |
| 21 | GPT-6.1 Sol - Provider Adapter (Medium) | OpenAI | 91.0 | 5834.91 |
| 22 | GPT-6.1 Sol - Provider Adapter (Low) | OpenAI | 82.8 | 6622.20 |
| 23 | Claude Fable 5.2 (XHigh) | Anthropic | 81.2 | 21948.83 |
| 24 | Claude Fable 5.2 (High) | Anthropic | 79.3 | 25173.94 |
| 25 | Claude Fable 5.2 (Medium) | Anthropic | 73.3 | 29034.48 |

Notes: ARC-AGI-3 semi-private set, from the leaderboard data file behind arcprize.org/leaderboard; raw fractions converted to % (1 decimal). Rows are system/effort configurations, several per model; 'Provider Adapter' entries use a provider-specific harness. Cost is as listed in the file (units per ARC Prize; shown to 2 decimals). ARC-AGI-1 and ARC-AGI-2 are near-saturated (top systems ~98.5% and ~95%) and are also in v1.json/v2.json.
