- MathVista 82.5
- MathVerse 81
- MathVision 61.9
- MMMU 72.2
- VLMs Are Blind 77.3
- VisuLogic 31.1
- VisualPuzzle 52.5
- ChartQA 87.1
- AI2D (transparent) 95.3
- DocVQA (val) 93.6
- OCRBench 858 (out of 1000) 85.8
- CharXiv-RQ (DQ 90.3) 56.9
- SimpleVQA 53.1
- HallusionBench 60.4
- MMStar 71.1
- BLINK 62.8
- CV-Bench 83.8
- CountBench 89
- TempCompass 75.1
ERNIE-4.5-VL-28B-A3B-Thinking
ERNIE-4.5-VL-28B-A3B-Thinking places in the top three on AI2D.
19
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
Baidu Nov 2025 ERNIE-4.5-VL-28B-A3B-Thinking
Open weights Small
No Epoch Capabilities Index score yet
- SimpleVQA 53.1% #24 /30
- HallusionBench 60.4% #22 /31
- BLINK 62.8% #17 /29
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HallusionBench | 60.4% | #22 of 31 | Lab model card | |
| SimpleVQA | 53.1% | #24 of 30 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AI2D | 95.3% | #3 of 52 | Lab model card | |
| MathVerse | 81% | #4 of 12 | Lab model card | |
| MathVista | 82.5% | #26 of 70 | Lab model card | |
| ChartQA | 87.1% | #17 of 37 | Lab model card | |
| MMMU | 72.2% | #55 of 103 | Lab model card | |
| DocVQA | 93.6% | #20 of 36 | Lab model card | |
| BLINK | 62.8% | #17 of 29 | Lab model card | |
| OCRBench | 85.8% | #21 of 34 | Lab model card | |
| CVBench | 83.8% | #10 of 15 | Lab model card | |
| CountBench | 89% | #20 of 28 | Lab model card | |
| MMStar | 71.1% | #26 of 36 | Lab model card | |
| MATH-Vision (MathVision) | 61.9% | #39 of 51 | Lab model card | |
| VLMs Are Blind | 77.3% | #12 of 14 | Lab model card | |
| TempCompass | 75.1% | #6 of 7 | Lab model card | |
| VisuLogic | 31.1% | #6 of 7 | Lab model card | |
| CharXiv | 56.9% | #66 of 75 | Lab model card | |
| VisualPuzzles | 52.5% | #3 of 3 | Lab model card |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.