What this page answers
A powerful multimodal Mixture-of-Experts chat model featuring 28B total parameters with 3B activated per token, delivering exceptional text and vision understanding through its innovative heterogeneous MoE structure with
- Baidu Wenxin · baidu/ernie-4.5-vl-28b-a3b
- text+image->text · China model route
- 30,000 context · $0.14 input
Before connecting
Do not stop at the model name. Before integration, verify base URL, protocol, visible models, parameters, and limits together.
- supports frequency_penalty
- supports include_reasoning
- supports max_tokens
- supports presence_penalty
- supports reasoning
Next action
The goal is to catch search demand, then move users into model profiles, provider profiles, and key checking.
- Check whether the model fits the use case
- Then verify key permission and callable models
