Horangi 4 - Final Score: leaderboard
Overall score of the Horangi Korean LLM Leaderboard 4 (W&B Korea): the mean of its GLP (general language performance) and ALT (alignment) averages over 24 Korean benchmarks, 0-100, higher is better. Runs scored before HumanEval-100 and BigCodeBench-100 joined count SWE-bench alone as Coding.
Metric: Score (%). Source: horangi.ai. Saturation forecast: Around February 2027. 105 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 84.14 | #64 (Gemini 3.1 Pro (Preview)) |
| 2 | Claude Fable 5 (High) | 82.62 | #20 (Claude Fable 5) |
| 3 | GPT-5.5 (xHigh) | 81.78 | #33 (GPT-5.5) |
| 4 | Gemini 3.5 Flash | 81.19 | #66 |
| 5 | Gemini 3 Pro (Preview) (High) | 80.76 | #71 (Gemini 3 Pro (Preview)) |
| 6 | GPT-5.6 Sol (Max) | 80.67 | #22 (GPT-5.6 Sol) |
| 7 | Qwen 3.7 Max | 80.2 | #85 |
| 8 | Claude Opus 4.6 (High) | 80.13 | #67 (Claude Opus 4.6) |
| 9 | GPT-5.4 (2026-03-05) (xHigh) | 79.7 | #63 (GPT-5.4 (2026-03-05)) |
| 10 | Grok 4.5 | 78.7 | #54 |
| 11 | Claude Sonnet 5 (High) | 78.64 | #65 (Claude Sonnet 5) |
| 12 | GPT-5.6 Terra (Max) | 78.31 | #40 (GPT-5.6 Terra) |
| 13 | Claude Opus 4.5 (20251101) (High) | 77.47 | #81 (Claude Opus 4.5 (20251101)) |
| 14 | GLM-5.2 (High) | 77.21 | #80 (GLM-5.2) |
| 15 | Qwen 3.7 Plus | 77.05 | #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-final-score · How It Works · Data refreshed daily, snapshot 2026-11-05.