kanana-2-30B-a3B (Thinking): benchmark results

Provider: Other.

Unified ELO 1447 ± 1, rank #1353 of 2032 rated models, from 36 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Horangi 4 - Ko-HalluLens (WikiQA)36Correct answer rate (%)68.8
Horangi 4 - Ko-HLE20Accuracy (%)64.4
Sonar LLM Leaderboard - Java - Vulnerability Density0.34Vulnerabilities per 1,000 lines of code (lower is better)29.3
Horangi 4 - GLP - Specialized Knowledge33Score (%)28.8
Horangi 4 - Ko-HellaSwag63Accuracy (%)19.7
Horangi 4 - SWE-bench Verified (80-task subset)3.75Resolved (%)19.2
Horangi 4 - KoBBQ68Accuracy (%)17.3
Horangi 4 - Ko-HalluLens (Nonexistent Entities)47Refusal rate (%)16.8
Horangi 4 - Ko-Moral59Accuracy (%)16.8
Horangi 4 - BigCodeBench (100-task subset)37Pass rate (%)16.3
Sonar LLM Leaderboard - Java - Code Smell Density22.33Code smells per 1,000 lines of code (lower is better)15.7
Horangi 4 - ALT - Hallucination Prevention45Score (%)14.4

Interactive version: theaggregate.ai/model?slug=kanana-2-30b-a3b-thinking · How It Works · Data refreshed daily, snapshot 2026-09-26.