kanana-2-30B-a3B-Thinking: benchmark results
Provider: Kakao. Released 2025-12-18. Access: Open.
Aggregate Rating 1451 ± 31, rank #3649 of 10870 rated models, from 36 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Horangi 4 - Ko-HalluLens (WikiQA) | 36 | Correct answer rate (%) | 68.8 |
| Horangi 4 - Ko-HLE | 20 | Accuracy (%) | 64.4 |
| Horangi 4 - GLP - Specialized Knowledge | 33 | Score (%) | 28.8 |
| Sonar LLM Leaderboard - Java - Vulnerability Density | 0.34 | Vulnerabilities per 1,000 lines of code (lower is better) | 26.7 |
| Horangi 4 - Ko-HellaSwag | 63 | Accuracy (%) | 19.7 |
| Horangi 4 - SWE-bench Verified (80-task subset) | 3.75 | Resolved (%) | 19.2 |
| Horangi 4 - KoBBQ | 68 | Accuracy (%) | 17.3 |
| Horangi 4 - Ko-HalluLens (Nonexistent Entities) | 47 | Refusal rate (%) | 16.8 |
| Horangi 4 - Ko-Moral | 59 | Accuracy (%) | 16.8 |
| Horangi 4 - BigCodeBench (100-task subset) | 37 | Pass rate (%) | 16.3 |
| Sonar LLM Leaderboard - Java - Code Smell Density | 22.33 | Code smells per 1,000 lines of code (lower is better) | 14.7 |
| Horangi 4 - ALT - Hallucination Prevention | 45 | Score (%) | 14.4 |
Interactive version: theaggregate.ai/model?slug=kanana-2-30b-a3b-thinking · How It Works · Data refreshed daily, snapshot 2026-11-12.