EXAONE 4.5 33B: benchmark results
LG AI Research's open 33B vision-language model with hybrid attention (research license). Provider: LG AI. Released 2026-04-09. Access: Open.
Aggregate Rating 1615 ± 18, rank #396 of 1656 rated models, from 98 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Horangi 4 - Ko-HalluLens (Nonexistent Entities) | 97 | Refusal rate (%) | 98.1 |
| K-MetBench | 75.9 | Accuracy (imported) | 81 |
| Horangi 4 - BFCL | 68.25 | Accuracy (%) | 77.5 |
| Horangi 4 - HumanEval (100-problem subset) | 98 | Pass rate (%) | 76.9 |
| Horangi 4 - Ko-Moral | 80 | Accuracy (%) | 74 |
| Horangi 4 - HRM8K | 93 | Accuracy (%) | 71.2 |
| Horangi 4 - IFEval-Ko | 87.92 | Instruction-following accuracy (%) | 70.2 |
| AA IFBench | 57.96 | Accuracy (%) | 69.8 |
| Tau2-Bench Telecom | 78.07 | Success Rate (%) | 69.6 |
| AA TAU-2 Bench | 78.07 | Accuracy (%) | 68.5 |
| AA GPQA Diamond | 79.39 | Accuracy (%) | 66.3 |
| LLM Stats (MathVista-Mini) | 85 | Score (%) | 64 |
Interactive version: theaggregate.ai/model?slug=exaone-4-5-33b · How It Works · Data refreshed daily, snapshot 2026-10-20.