reasoning benchmark
ARC-AGI v2 Leaderboard
ARC-AGI v2 is a benchmark for evaluating AI systems on abstract reasoning and problem-solving through visual grid transformation tasks.
Updated Oct 23, 2026
Models19
Model coverage19
MetricScore
EvidenceB
ARC-AGI v2 Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelOP | Score95.00% | Percentile100.00% | Participants19 | EvidenceC | Evaluated |
| Rank02 | ModelOP | Score85.00% | Percentile94.44% | Participants19 | EvidenceC | Evaluated |
| Rank03 | ModelGO | Score77.10% | Percentile88.89% | Participants19 | EvidenceC | Evaluated |
| Rank04 | ModelOP | Score73.30% | Percentile83.33% | Participants19 | EvidenceC | Evaluated |
| Rank05 | ModelGO | Score72.10% | Percentile77.78% | Participants19 | EvidenceC | Evaluated |
| Rank06 | ModelAN | Score68.80% | Percentile72.22% | Participants19 | EvidenceC | Evaluated |
| Rank07 | ModelAN | Score58.30% | Percentile66.67% | Participants19 | EvidenceC | Evaluated |
| Rank08 | ModelOP | Score54.20% | Percentile61.11% | Participants19 | EvidenceC | Evaluated |
| Rank09 | ModelOP | Score52.90% | Percentile55.56% | Participants19 | EvidenceC | Evaluated |
| Rank10 | ModelME | Score42.50% | Percentile50.00% | Participants19 | EvidenceC | Evaluated |
| Rank11 | ModelTM | Score40.10% | Percentile44.44% | Participants19 | EvidenceC | Evaluated |
| Rank12 | ModelAN | Score37.60% | Percentile38.89% | Participants19 | EvidenceC | Evaluated |
| Rank13 | ModelGO | Score33.60% | Percentile33.33% | Participants19 | EvidenceC | Evaluated |
| Rank14 | ModelGO | Score31.10% | Percentile27.78% | Participants19 | EvidenceC | Evaluated |
| Rank15 | ModelXA | Score15.90% | Percentile22.22% | Participants19 | EvidenceC | Evaluated |
| Rank16 | ModelAN | Score8.60% | Percentile16.67% | Participants19 | EvidenceB | Evaluated |
| Rank17 | ModelOP | Score6.50% | Percentile11.11% | Participants19 | EvidenceB | Evaluated |
| Rank18 | ModelGO | Score4.90% | Percentile5.56% | Participants19 | EvidenceB | Evaluated |
| Rank19 | ModelBY | Score1.50% | Percentile0.00% | Participants19 | EvidenceC | Evaluated |
ARC-AGI v2 Highlights
The leading models and scores on this benchmark.
ARC-AGI v2 Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for ARC-AGI v2
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is ARC-AGI v2?
What ARC-AGI v2 measures and how its scores work.
It presents input-output grid pairs with colored cells and requires models to identify transformation rules from demonstration examples and apply them to test cases.
It measures abstract reasoning, problem-solving, fluid intelligence, spatial reasoning, pattern recognition, and compositional generalization using the Score metric in ratio units.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- ARC-AGI v2
- Modality
- multimodal
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about ARC-AGI v2.
Which model scores highest on ARC-AGI v2?
GPT-6 Astra is currently ranked first with 95.00%.
What are the top three models on ARC-AGI v2?
The current leaders are GPT-6 Astra (95.00%), GPT-5.5 (85.00%), and Gemini 3.1 Pro (77.10%).
Which ARC-AGI v2 model has the lowest official input price?
Gemini 3 Flash has the lowest matched official input price at $0.50 input / $3.0 output per 1M tokens.
Which models are fastest among ARC-AGI v2 results?
The fastest matched records are GPT-5.2 (100.00 tok/s via OpenAI), Grok-4 (100.00 tok/s via xAI), and Claude Opus 4 (100.00 tok/s via Anthropic).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.