activeReasoning

ARC-AGI-2

Metric: Score (higher is better)

Second iteration of the Abstraction and Reasoning Corpus for AGI evaluation

Models ranked

27

tracked on this benchmark

Score band

95.0 – 10.3

best → lowest tracked

Snapshot trend

-5.06

Jun 7 → Oct 1 · 17 models

Leaderboard

Tracked models ranked by Score (higher is better).

Compare candidates
#Model variant and provenanceScore
1
GPT-6 Astra

Configuration: gpt-6-astra (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'GPT-6 Astra (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

95.0
2
GPT-6.1 Sol

Configuration: gpt-6.1-sol (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'GPT-6.1 Sol (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

94.2
3
Claude Opus 5.5

Configuration: claude-opus-5-5 (High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Claude Opus 5.5 (High)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

93.3
4
Claude Fable 5.1

Configuration: claude-fable-5-1 (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Claude Fable 5.1 (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

90.0
5
GPT-6 Sol

Configuration: gpt-6-sol (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'GPT-6 Sol (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

89.6
6
Gemini 3.8 Flash

Configuration: gemini-3.8-flash (High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Gemini 3.8 Flash (High)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

89.2
7
GPT-5.5
Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
85.0
8
Gemini 3.7 Flash

Configuration: gemini-3.7-flash (High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Gemini 3.7 Flash (High)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

84.6
9
Gemini 3.1 Pro Preview
Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
77.1
10
Claude Opus 4.7
Version: Listed as 'Claude Opus 4 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
75.8
11
GPT-5.4
Version: llm-stats shows 0 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
73.3
12
Gemini 3.5 Flash
Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
72.1
13
Claude Opus 4.6
Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
68.8
14
Grok 4.6

Configuration: grok-4.6 (Extra High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Grok 4.6 (XHigh)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

67.1
15
GLM-5.3-Flash

Configuration: glm-5.3-flash (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'GLM-5.3-Flash (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

65.8
16
Gemini 3.6 Flash

Configuration: gemini-3.6-flash (High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Gemini 3.6 Flash (High)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

60.4
17
Kimi K3

Configuration: kimi-k3 (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Kimi K3 (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

60.4
18
GPT-6 Luna

Configuration: gpt-6-luna (Max)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'GPT-6 Luna (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

59.3
19
Claude Sonnet 4.6
Version: llm-stats shows 0 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
58.3
20
Grok 4.20
Version: ARC-AGI-2 (accuracy%)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
53.3
21
Qwen3.8-27B

Configuration: qwen3.8-27b (Extra High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Qwen3.8-27B (XHigh)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

42.4
22
Inkling-Small

Configuration: Inkling-Small (XHigh)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official data https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01) modelDisplayName 'Inkling Small (XHigh)' score 0.40139; matches Epoch AI mirror. Highest published effort.

40.1
23
Claude Opus 4.5
Version: Not recordedHarness: Not recordedEvaluator: Not recordedObserved: Nov 1, 2025Confidence: Not recordedSource
37.6
24
Inkling

Configuration: Inkling (single reported configuration)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data, verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) modelDisplayName 'Inkling' score 0.36528 (no effort label published). Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

36.5
25
Gemini 3 Flash Preview

Configuration: gemini-3-flash-preview (High)

Version: ARC-AGI-2 semi-privateHarness: ARC Prize semi-private evaluationEvaluator: ARC Prize FoundationObserved: Oct 1, 2026Confidence: confirmedSource

Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Gemini 3 Flash Preview (High)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only).

33.6
Showing the top 25 of 27 tracked models. Browse all models.

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

GPQA
Google-Proof Q&A
DROP
Discrete Reasoning Over Paragraphs
ARC
AI2 Reasoning Challenge
HellaSwag
MMMU Pro
ANLS*
Average Normalized Levenshtein Similarity*

Last reviewed: May 20, 2026

Resources

Website