activeReasoning
ARC-AGI-2
Metric: Score (higher is better)
Second iteration of the Abstraction and Reasoning Corpus for AGI evaluation
Models ranked
10
tracked on this benchmark
Score band
85.0 – 15.9
best → lowest tracked
Snapshot trend
-21.70
Nov 1 → Jul 1 · 1 models
Leaderboard
Tracked models ranked by Score (higher is better).
#Model variant and provenanceRelative to leaderScore
1
GPT-5.5
85.0Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
2
Gemini 3.1 Pro Preview
77.1Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
3
Claude Opus 4.7
75.8Version: Listed as 'Claude Opus 4 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
4
GPT-5.4
73.3Version: llm-stats shows 0 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
5
Gemini 3.5 Flash
72.1Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
6
Claude Opus 4.6
68.8Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
7
Claude Sonnet 4.6
58.3Version: llm-stats shows 0 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
8
Grok 4.20
53.3Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
9
Claude Opus 4.5
37.6Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Nov 1, 2025Confidence: Not recordedSource
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: May 20, 2026