Reasoning benchmark
ARC-AGI-2 leaderboard
Abstraction and Reasoning Corpus for AGI v2. Every model the catalog carries a published ARC-AGI-2 value for, ranked by that value.
A benchmark measuring fluid intelligence and novel abstract reasoning through visual grid puzzles. Models must identify patterns in input-output pairs and generate the correct output for unseen inputs. Considered the hardest public reasoning benchmark — average individual human performance is 66%.
ARC-AGI-2 ranking
22 models with a published ARC-AGI-2 value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Grid transformation puzzles with novel rules |
|---|---|---|---|
| 1 |  GPT-6 Astra | OpenAI | 95 |
| 2 | GPT-5.6 Sol | OpenAI | 92.5 |
| 3 |  Claude Opus 5 | Anthropic | 90.4 |
| 4 | Claude Fable 5.1 | Anthropic | 90 |
| 5 | GPT-5.5 | OpenAI | 85 |
| 6 | GPT-5.6 Terra | OpenAI | 83.9 |
| 7 | GPT-5.4 Pro | OpenAI | 83.3 |
| 8 | DSdots3-note Preview | Dots Studio | 81.4 |
| 9 |  Gemini 3.1 Pro | 77.08 | |
| 10 | Claude Opus 4.7 (Adaptive) | Anthropic | 75.8 |
| 11 | GPT-5.4 | OpenAI | 73.95 |
| 12 | Gemini 3.5 Flash | 72.1 | |
| 13 | Claude Opus 4.8 | Anthropic | 72.08 |
| 14 | GPT-5.6 Luna | OpenAI | 59.54 |
| 15 |  Grok 4.20 | xAI | 53.3 |
| 16 | GPT-5.2 | OpenAI | 52.9 |
| 17 | Grok 4.5 | xAI | 52.64 |
| 18 | Gemini 3 Pro Deep Think | 45.1 | |
| 19 |  Muse Spark | Meta | 42.5 |
| 20 | TMInkling-Small | Thinking Machines Lab | 40.1 |
| 21 | Gemini 3 Pro | 31.1 | |
| 22 | Claude Sonnet 4.5 | Anthropic | 13.6 |
Evidence key: Observed
Rows are ordered by the value ARC-AGI-2: A Harder General Intelligence Benchmark published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.