ERNIE 4.5 VL 28B A3B

An estimated 21 GB at Q4, 36.4 GB at Q8 and 68.1 GB at FP16, based on the 29.4B parameter count. All VRAM figures are estimates, covering weights plus about 15% and 0.5 GB for a short context.

Yes on a 24 GB GPU, which fits at Q4 with an estimated 21 GB. A 16 GB, 12 GB or 8 GB GPU does not fit at Q4, Q8 or FP16, per estimates.

Yes, with an estimated 21 GB needed at Q4. Assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits is 32 GB, an estimate.

Yes. The model has open weights, so you can self-host it for full control of your data.

Run ERNIE 4.5 VL 28B A3B locally

ERNIE 4.5 VL 28B A3B accepts text and image inputs. Audio, video and PDFs are not listed.

Matched on model kind and release date, with no comparable price, the closest alternatives are DiffuCoder 7B by Apple, Jamba Large by AI21 Labs and Gemma 3n E2B IT by Google DeepMind.

DiffuCoder 7BERNIE 4.5 VL 28B A3B vs DiffuCoder 7BJamba LargeERNIE 4.5 VL 28B A3B vs Jamba LargeGemma 3n E2B ITERNIE 4.5 VL 28B A3B vs Gemma 3n E2B IT

ERNIE 4.5 VL 28B A3B has open weights. The license is Apache 2.0.

Yes. The license is Apache 2.0, a permissive license that allows commercial use.

From the Hugging Face repository baidu/ERNIE-4.5-VL-28B-A3B-PT.

Run ERNIE 4.5 VL 28B A3B locally

ERNIE 4.5 VL 28B A3B was made by Baidu and released on June 30, 2025.

More from Baidu

Yes, the model has open weights.