Back to Models
BaiduOpen Source

ERNIE-4.5-VL-28B-A3B Multimodal MoE Chat Model

ERNIE-4.5-VL-28B-A3B is a multi-modal foundation model developed by Baidu. It adopts a MoE architecture, featuring a vast parameter count and a long 131K context window.

Parameters

280.0B

Context Window

131K

License

Apache 2.0

Release Date

2025-08-20

API Pricing

API pricing for this model is not yet available

Strengths

  • Large parameter count of 280B
  • Supports long 131K context
  • Uses efficient MoE architecture

Weaknesses

  • High computational load from model size
  • Intensive resource consumption
  • Requires advanced hardware

Use Cases

  • Complex multimodal analysis
  • Understanding and processing long contexts
  • Advanced vision-language integrated chat

Deep Analysis

Parameters

28B total, 3B activated

Context Window

128K tokens

Pricing

$0.14/$0.56 per 1M tokens

License

Apache 2.0

Release Date

August 12, 2025

Strengths

  • Efficient MoE with 3B activated params
  • Thinking/non-thinking modes
  • Vision capabilities
  • Affordable

Weaknesses

  • Limited to 30K on APIs
  • No video/audio input
  • Lower scores vs larger models

Competitor Comparison

Model
Qwen3-VL
Llama 4 Scout

ERNIE-4.5-VL-28B-A3B is a multimodal MoE model with 28B/3B activated params. Heterogeneous MoE with modality-isolated routing.

Analysis generated: 2026-05-24