ERNIE-4.5-VL-424B-A47B: benchmark results
Baidu's ERNIE 4.5 VL, a 424B (47B active) multimodal mixture-of-experts model jointly pre-trained on text and images with modality-isolated routing for cross-modal reasoning. Provider: Baidu. Released 2025-06-30. Access: Open.
Aggregate Rating 1617 ± 20, rank #381 of 1635 rated models, from 89 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - ocw | 1.83 | Dataset z-score | 100 |
| FlagEval VQA - Text Understanding (Chinese) | 80.3 | Score | 100 |
| AGC-Bench - tinystories | 1.37 | Dataset z-score | 93.9 |
| AGC-Bench - mops | 1.26 | Dataset z-score | 91.5 |
| AGC-Bench - cue_word_story | 0.96 | Dataset z-score | 90.9 |
| AGC-Bench - irfl | 1.21 | Dataset z-score | 90.6 |
| AGC-Bench - unfun_corpus | 0.63 | Dataset z-score | 89.7 |
| AGC-Bench - riddlesense | 0.95 | Dataset z-score | 87.5 |
| AGC-Bench - fann_or_flop | 0.95 | Dataset z-score | 86.6 |
| AGC-Bench - schnovel | 1.06 | Dataset z-score | 84.1 |
| AGC-Bench - moh_x | 0.76 | Dataset z-score | 84 |
| AGC-Bench - proparalogy | 0.77 | Dataset z-score | 82.1 |
Interactive version: theaggregate.ai/model?slug=ernie-4-5-vl-424b-a47b · How It Works · Data refreshed daily, snapshot 2026-10-15.