reasoning benchmark
ARC-AGI v2 Leaderboard
ARC-AGI-2 is an upgraded benchmark for measuring abstract reasoning and problem-solving abilities in AI systems through visual grid transformation tasks. It evaluates fluid intelligence via input-output grid pairs (1x1 to 30x30) using colored cells (0-9), requiring models to identify underlying transformation rules from demonstration examples and apply them to test cases. Designed to be easy for humans but challenging for AI, focusing on core cognitive abilities like spatial reasoning, pattern recognition, and compositional generalization.
Updated Aug 17, 2026
Models17
Model coverage17
MetricScore
EvidenceB
ARC-AGI v2 Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelOP | Score85.0% | Percentile100.0% | Participants17 | EvidenceC | Evaluated |
| Rank02 | ModelGO | Score77.1% | Percentile93.8% | Participants17 | EvidenceC | Evaluated |
| Rank03 | ModelOP | Score73.3% | Percentile87.5% | Participants17 | EvidenceC | Evaluated |
| Rank04 | ModelGO | Score72.1% | Percentile81.3% | Participants17 | EvidenceC | Evaluated |
| Rank05 | ModelAN | Score68.8% | Percentile75.0% | Participants17 | EvidenceC | Evaluated |
| Rank06 | ModelAN | Score58.3% | Percentile68.8% | Participants17 | EvidenceC | Evaluated |
| Rank07 | ModelOP | Score54.2% | Percentile62.5% | Participants17 | EvidenceC | Evaluated |
| Rank08 | ModelOP | Score52.9% | Percentile56.3% | Participants17 | EvidenceC | Evaluated |
| Rank09 | ModelME | Score42.5% | Percentile50.0% | Participants17 | EvidenceC | Evaluated |
| Rank10 | ModelTM | Score40.1% | Percentile43.8% | Participants17 | EvidenceC | Evaluated |
| Rank11 | ModelAN | Score37.6% | Percentile37.5% | Participants17 | EvidenceC | Evaluated |
| Rank12 | ModelGO | Score33.6% | Percentile31.3% | Participants17 | EvidenceC | Evaluated |
| Rank13 | ModelGO | Score31.1% | Percentile25.0% | Participants17 | EvidenceC | Evaluated |
| Rank14 | ModelXA | Score15.9% | Percentile18.8% | Participants17 | EvidenceC | Evaluated |
| Rank15 | ModelAN | Score8.6% | Percentile12.5% | Participants17 | EvidenceB | Evaluated |
| Rank16 | ModelOP | Score6.5% | Percentile6.3% | Participants17 | EvidenceB | Evaluated |
| Rank17 | ModelGO | Score4.9% | Percentile0.0% | Participants17 | EvidenceB | 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.
Ranking basisThis arc-agi v2 AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01OPGPT-5.5OpenAI
- Score
- 85.0%
- Price
- $5.0 input / $30 output per 1M tokens
Strengths
- Ranks #1 of 17 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures ARC-AGI v2, not total model capability
- 02GOGemini 3.1 ProGoogle
- Score
- 77.1%
- Price
- $2.0 input / $12 output per 1M tokens
- Speed
- Up to 23 tok/s via Google
Strengths
- Ranks #2 of 17 compared models
- 94th percentile on this benchmark
- C evidence result
Considerations
- This result measures ARC-AGI v2, not total model capability
- 03OPGPT-5.4OpenAI
- Score
- 73.3%
- Price
- $2.5 input / $15 output per 1M tokens
- Speed
- Up to 9.1 tok/s via OpenAI
Strengths
- Ranks #3 of 17 compared models
- 88th percentile on this benchmark
- C evidence result
Considerations
- This result measures ARC-AGI v2, not total model capability
- 04GOGemini 3.5 FlashGoogle
- Score
- 72.1%
- Price
- $1.5 input / $9.0 output per 1M tokens
- Speed
- Up to 1.2 tok/s via Google
Strengths
- Ranks #4 of 17 compared models
- 81th percentile on this benchmark
- C evidence result
Considerations
- This result measures ARC-AGI v2, not total model capability
- 05ANClaude Opus 4.6Anthropic
- Score
- 68.8%
- Price
- $5.0 input / $25 output per 1M tokens
- Speed
- Up to 17 tok/s via Anthropic
Strengths
- Ranks #5 of 17 compared models
- 75th percentile on this benchmark
- C evidence result
Considerations
- This result measures ARC-AGI v2, not total model capability
Selection summary
Best AI Models for ARC-AGI v2
GPT-5.5 currently leads ARC-AGI v2 with 85.0%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is ARC-AGI v2?
What ARC-AGI v2 measures and how its scores work.
ARC-AGI-2 is an upgraded benchmark for measuring abstract reasoning and problem-solving abilities in AI systems through visual grid transformation tasks. It evaluates fluid intelligence via input-output grid pairs (1x1 to 30x30) using colored cells (0-9), requiring models to identify underlying transformation rules from demonstration examples and apply them to test cases. Designed to be easy for humans but challenging for AI, focusing on core cognitive abilities like spatial reasoning, pattern recognition, and compositional generalization.
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
- Yes
- Evaluation key
- arc-agi-v2|llm-stats-current
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-5.5 is currently ranked first with 85.0%.
What does ARC-AGI v2 measure?
ARC-AGI-2 is an upgraded benchmark for measuring abstract reasoning and problem-solving abilities in AI systems through visual grid transformation tasks. It evaluates fluid intelligence via input-output grid pairs (1x1 to 30x30) using colored cells (0-9), requiring models to identify underlying transformation rules from demonstration examples and apply them to test cases. Designed to be easy for humans but challenging for AI, focusing on core cognitive abilities like spatial reasoning, pattern recognition, and compositional generalization.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
17 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.