ARC-AGI-1
The original Abstraction and Reasoning Corpus — grid puzzles solved from a handful of examples. Semi-private evaluation set, verified by ARC Prize.
Hard, novel problems: graduate science, abstract puzzles, expert exams. One of 6 comparable tests in reasoning, a category the Index averages.
Read with care. Close to saturated at the frontier; scores depend on the compute budget a lab chose.
Publisher: ARC Prize FoundationWhat a model is asked to do
Solve the original Abstraction and Reasoning Corpus: small grids, a handful of examples, one hidden rule.
For exampleTwo examples show a pattern being extended along its row until it reaches the border. Complete the third grid.
Why it matters. The first ARC set is close to saturated at the frontier, so it now shows whether smaller models can generalize at all.
The example is original and illustrative, not an item from the dataset.
- Models scored here
- 51
- Who produced the numbers
- Official leaderboard
- Items graded
- n = 100 (1.0% each)
- Best published result
- 98.5%
- License
- CC BY 4.0 (via Epoch AI)
- Direction
- Higher is better
- Scale
- Published as a share of items solved. For the Index, each score is a share of the best published result (98.5%), so the frontier reads 100. Guessing earns nothing on this test, so no chance correction applies.
- Provenance
- Produced by the benchmark's own leaderboard or a third-party evaluator, not by the model's maker.
Snapshot September 8, 2026
Ranking on this test
51 models on ARC-AGI-1
Best published run per model, and how close each one comes to the best published result on this test (the frontier reads 100). Every number links to the publisher that produced it. Internal runs are marked and carry their run details on the model page. One item on this test is 1.0 points, so read gaps smaller than that as noise.
| # | Model | Score | vs frontier | Run | Produced by |
|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 98.5% | 100.0 | xhigh | Leaderboard |
| 2 | Claude Fable 5Anthropic | 98.5% | 100.0 | xhigh | Leaderboard |
| 3 | Gemini 3.1 ProGoogle | 98.0% | 99.5 | preview | Leaderboard |
| 4 | Claude Fable 5.1Anthropic | 97.5% | 99.0 | max | Leaderboard |
| 5 | Claude Opus 5Anthropic | 97.5% | 99.0 | max | Leaderboard |
| 6 | GPT-5.6 SolOpenAI | 97.5% | 99.0 | xhigh | Leaderboard |
| 7 | GPT-5.6 TerraOpenAI | 96.5% | 98.0 | max | Leaderboard |
| 8 | GPT-5.5 ProOpenAI | 96.5% | 98.0 | high | Leaderboard |
| 9 | Gemini 3 Deep ThinkGoogle | 96.0% | 97.5 | preview | Leaderboard |
| 10 | Gemini 3.7 FlashGoogle | 95.5% | 97.0 | high | Leaderboard |
| 11 | GPT-5.5OpenAI | 95.0% | 96.4 | xhigh | Leaderboard |
| 12 | Kimi K3Moonshot AI | 94.5% | 95.9 | max | Leaderboard |
| 13 | GPT-5.4 ProOpenAI | 94.5% | 95.9 | 2026-03-05 · xhigh | Leaderboard |
| 14 | Claude Opus 4.6Anthropic | 94.0% | 95.4 | 120k | Leaderboard |
| 15 | GPT-5.4OpenAI | 93.7% | 95.1 | 2026-03-05 · xhigh | Leaderboard |
| 16 | Claude Opus 4.7Anthropic | 93.5% | 94.9 | high | Leaderboard |
| 17 | Claude Opus 4.8Anthropic | 92.5% | 93.9 | max | Leaderboard |
| 18 | Gemini 3.5 FlashGoogle | 92.5% | 93.9 | high | Leaderboard |
| 19 | Gemini 3.6 FlashGoogle | 91.2% | 92.6 | high | Leaderboard |
| 20 | DeepSeek V4 ProDeepSeek | 90.5% | 91.9 | low | Leaderboard |
| 21 | GPT-5.2 ProOpenAI | 90.5% | 91.9 | 2025-12-11 · xhigh | Leaderboard |
| 22 | Grok 4.20xAI | 89.5% | 90.9 | — | Leaderboard |
| 23 | DeepSeek V4 FlashDeepSeek | 89.0% | 90.4 | max | Leaderboard |
| 24 | GPT-5.6 LunaOpenAI | 88.0% | 89.3 | max | Leaderboard |
| 25 | Grok 4.6xAI | 87.5% | 88.8 | medium | Leaderboard |
| 26 | Grok 4.5xAI | 87.2% | 88.5 | medium | Leaderboard |
| 27 | Claude Sonnet 4.6Anthropic | 86.5% | 87.8 | high | Leaderboard |
| 28 | GPT-5.2OpenAI | 86.2% | 87.5 | 2025-12-11 · xhigh | Leaderboard |
| 29 | Gemini 3 FlashGoogle | 84.7% | 86.0 | preview | Leaderboard |
| 30 | Inkling SmallThinking Machines | 84.0% | 85.3 | xhigh | Leaderboard |
| 31 | Claude Opus 4.5Anthropic | 80.0% | 81.2 | 20251101 · 64k | Leaderboard |
| 32 | InklingThinking Machines | 79.5% | 80.7 | — | Leaderboard |
| 33 | GLM 5.2Zhipu AI | 77.0% | 78.2 | — | Leaderboard |
| 34 | Gemini 3 ProGoogle | 75.0% | 76.1 | preview | Leaderboard |
| 35 | GPT-5.1OpenAI | 72.8% | 73.9 | 2025-11-13 · high | Leaderboard |
| 36 | GPT-5 ProOpenAI | 70.2% | 71.3 | 2025-10-06 · high | Leaderboard |
| 37 | Grok 4xAI | 66.7% | 67.7 | — | Leaderboard |
| 38 | GPT-5OpenAI | 65.7% | 66.7 | 2025-08-07 · high | Leaderboard |
| 39 | Kimi K2.5Moonshot AI | 65.3% | 66.3 | — | Leaderboard |
| 40 | Claude Sonnet 4.5Anthropic | 63.7% | 64.6 | 20250929 · 32k | Leaderboard |
| 41 | GPT-5.4 miniOpenAI | 63.7% | 64.6 | 2026-03-17 · xhigh | Leaderboard |
| 42 | MiniMax M2.5MiniMax | 63.7% | 64.6 | — | Leaderboard |
| 43 | DeepSeek V3.2DeepSeek | 57.0% | 57.9 | — | Leaderboard |
| 44 | GPT-5 miniOpenAI | 54.3% | 55.2 | 2025-08-07 · high | Leaderboard |
| 45 | Gemini 3.5 Flash-LiteGoogle | 53.5% | 54.3 | high | Leaderboard |
| 46 | GPT-5.4 nanoOpenAI | 51.5% | 52.3 | 2026-03-17 · xhigh | Leaderboard |
| 47 | Grok 4 FastxAI | 48.5% | 49.2 | — | Leaderboard |
| 48 | Claude Haiku 4.5Anthropic | 47.7% | 48.4 | 20251001 · 32k | Leaderboard |
| 49 | GLM 5Zhipu AI | 44.7% | 45.4 | — | Leaderboard |
| 50 | Gemini 2.5 ProGoogle | 41.0% | 41.6 | 16k | Leaderboard |
| 51 | GPT-5 nanoOpenAI | 20.7% | 21.0 | 2025-08-07 · medium | Leaderboard |
The best models, ranked here, working inside Theo.
28 of the models on this page run inside Theo today. Theo picks the right one for each step and always shows which engine answered.