Baidu: ERNIE 4.5 VL 28B A3B

Baidu: ERNIE 4.5 VL 28B A3B is a cutting-edge multimodal Mixture-of-Experts (MoE) chat model, boasting 28 billion total parameters with 3 billion activated per token. Its innovative heterogeneous MoE structure with modality-isolated routing ensures superior text and vision understanding, making it highly efficient for complex tasks. Built on a scaling-efficient infrastructure, ERNIE 4.5 VL 28B A3B supports high-throughput training and inference. It leverages advanced post-training techniques including SFT, DPO, and UPO for optimized performance. With an impressive 131K context length and RLVR alignment, it excels in cross-modal reasoning and generation. It offers vision, functions, and streaming capabilities. Context window: 30K tokens, Max output: 4K tokens. Pricing: $0.14/0.56 per 1M tokens (input/output).