Keye-VL-1.5-8B: benchmark results
Provider: Kuaishou. Released 2025-08-27. Access: Open.
Aggregate Rating 1555 ± 15, rank #1535 of 10870 rated models, from 106 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| E-VAds (Relaxed-5) - Cross-Modal Detection | 75.6 | Relaxed-5 score (%) on the Cross-Modal Detection task (consi | 100 |
| TempCompass | 75.53 | Avg All (%) | 100 |
| E-VAds (Relaxed-5) - Regulatory Compliance | 69.5 | Relaxed-5 score (%) on the Regulatory Compliance task (poten | 93.8 |
| Open Portuguese LLM - BLUEX | 65.92 | Accuracy (%) | 87.7 |
| Open Portuguese LLM - FaQuAD NLI | 79.18 | Macro F1 (%) | 86.6 |
| Open Portuguese LLM - ENEM | 73.2 | Accuracy (%) | 84.5 |
| E-VAds | 20.6 | Strict score (%): share of answers the judge rates fully cor | 81.2 |
| Open Portuguese LLM - OAB Exams | 51.94 | Accuracy (%) | 80.9 |
| Open Portuguese LLM - ASSIN2 RTE | 92.5 | Macro F1 (%) | 79.8 |
| VLM-DeflectionBench - Oracle Accuracy | 63.4 | Accuracy (%) of knowledge-based visual question answering in | 72.2 |
| VLM-DeflectionBench - Realistic Accuracy | 47.6 | Accuracy (%) of knowledge-based visual question answering in | 72.2 |
| SciVQR - Mathematics | 61.1 | Mathematics accuracy (%) with zero-shot chain-of-thought pro | 71.4 |
Interactive version: theaggregate.ai/model?slug=keye-vl-1-5-8b · How It Works · Data refreshed daily, snapshot 2026-11-12.