ERNIE 4.5 VL 28B A3B: benchmark results
Provider: Baidu. Released 2025-06-30. Access: API.
Aggregate Rating 1468 ± 23, rank #1206 of 2170 rated models, from 33 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SWE-Arena | 1002 | Elo Score | 81.9 |
| GEB-Bench - Story-Theorem Matching | 41.3 | Accuracy (%; 4-way, chance 25%) | 52.8 |
| CFMME - Information Extraction | 80.71 | Field-level F1 (%) of key information extracted from financi | 50 |
| CFMME - Seal Recognition | 73.86 | Seal text recognition score (%, 1 minus normalized edit dist | 50 |
| GEB-Bench - Scene-Story Matching | 32.4 | Accuracy (%; 4-way, chance 25%) | 46.2 |
| JP-TL-Bench - Japanese to English - Easy | 6.12 | LT Score (0-10; Easy items only; anchored pairwise Bradley-T | 38.1 |
| JP-TL-Bench - Japanese to English - Hard | 5.87 | LT Score (0-10; Hard items only; anchored pairwise Bradley-T | 38.1 |
| CFMME - Application QA | 42.07 | Accuracy (%) on question answering over real financial docum | 37.5 |
| CFMME - Table Recognition | 57.49 | Table recognition TEDS (%, tree-edit-distance similarity of | 37.5 |
| JP-TL-Bench - Japanese to English | 5.94 | LT Score (0-10; anchored pairwise Bradley-Terry vs 20 Base S | 36.5 |
| PhysCount - Correspondence NAE | 58.93 | Correspondence normalized absolute error (%) | 33.3 |
| GEB-Bench - Skeleton Identification | 46.9 | Accuracy (%; 25-way, chance 4%) | 30.8 |
Interactive version: theaggregate.ai/model?slug=ernie-4-5-vl-28b-a3b · How It Works · Data refreshed daily, snapshot 2026-11-05.