Horangi 4 - Ko-ARC-AGI: leaderboard
ARC-AGI abstract grid-reasoning puzzles with Korean prompts, as run by the Horangi Korean LLM Leaderboard 4 (W&B Korea). Accuracy (%) of exact grid matches, higher is better.
Metric: Accuracy (%). Source: horangi.ai. Saturation forecast: Around February 2027. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 97 | #22 (GPT-5.6 Sol) |
| 2 | GPT-5.6 Terra (Max) | 97 | #40 (GPT-5.6 Terra) |
| 3 | Claude Fable 5 (High) | 96 | #20 (Claude Fable 5) |
| 4 | GPT-5.4 (2026-03-05) (xHigh) | 95 | #63 (GPT-5.4 (2026-03-05)) |
| 5 | GPT-5.5 (xHigh) | 93 | #33 (GPT-5.5) |
| 6 | GPT-5.2 (2025-12-11) (xHigh) | 92.71 | |
| 7 | Gemini 3.1 Pro (Preview) (High) | 92 | #64 (Gemini 3.1 Pro (Preview)) |
| 8 | Motif 3 | 90.91 | #195 |
| 9 | Grok 4.20 | 90.72 | #179 |
| 10 | Gemini 3.5 Flash | 89.9 | #66 |
| 11 | Claude Sonnet 5 (High) | 85 | #65 (Claude Sonnet 5) |
| 12 | Grok 4.5 | 84.04 | #54 |
| 13 | Claude Opus 4.6 (High) | 84 | #67 (Claude Opus 4.6) |
| 14 | GPT-5.6 Luna (Max) | 82 | #79 (GPT-5.6 Luna) |
| 15 | Qwen 3.7 Plus | 80.46 | #98 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=horangi-4-ko-arc-agi · How It Works · Data refreshed daily, snapshot 2026-11-05.