HyperCLOVA X SEED Think (32B): benchmark results
Naver's open 32B Korean-centric vision-language reasoning model. Provider: Naver. Released 2025-12-26. Access: Open.
Aggregate Rating 1536 ± 21, rank #653 of 1635 rated models, from 88 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA TAU-2 Bench | 87.43 | Accuracy (%) | 81.4 |
| Tau2-Bench Telecom | 87.43 | Success Rate (%) | 81.2 |
| Horangi 4 - Ko-Moral | 81 | Accuracy (%) | 79.3 |
| AA Omniscience - Software Engineering (SWE) - HTML | 38 | Accuracy (%) | 72.7 |
| AA LiveCodeBench | 62.86 | Pass@1 (%) | 69 |
| AA MMLU-Pro | 78.49 | Accuracy (%) | 60.9 |
| Horangi 4 - BFCL | 67.44 | Accuracy (%) | 56.9 |
| AA Omniscience - Software Engineering (SWE) - Go | 18 | Accuracy (%) | 56.5 |
| AA AIME 2025 | 59 | Accuracy (%) | 55.9 |
| Horangi 4 - HAE-RAE Bench (Reading Comprehension) | 88 | Accuracy (%) | 53.4 |
| Horangi 4 - HAE-RAE Bench (without Reading Comprehension) | 86.87 | Accuracy (%) | 52.4 |
| AA Omniscience - Software Engineering (SWE) - Java | 15 | Accuracy (%) | 49.8 |
Interactive version: theaggregate.ai/model?slug=hyperclova-x-seed-think-32b · How It Works · Data refreshed daily, snapshot 2026-10-15.