Baidu: ERNIE 4.5 VL 28B A3B

Text input Image input Text output
Author's Description

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 modality-isolated routing. Built with scaling-efficient infrastructure for high-throughput training and inference, the model leverages advanced post-training techniques including SFT, DPO, and UPO for optimized performance, while supporting an impressive 131K context length and RLVR alignment for superior cross-modal reasoning and generation capabilities.

Key Specifications
Cost
$$$
Context
30K
Parameters
28B
Released
Aug 12, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Logit Bias Reasoning Include Reasoning Stop Seed Min P Top P Max Tokens Frequency Penalty Temperature Presence Penalty
Features

This model supports the following features:

Reasoning
Performance Summary

Baidu's ERNIE 4.5 VL 28B A3B, a powerful multimodal Mixture-of-Experts model, demonstrates competitive response times, ranking in the 52nd percentile across seven benchmarks. It also offers cost-effective solutions, placing in the 62nd percentile for price. A standout feature is its exceptional reliability, achieving a perfect 100% success rate across all benchmarks, indicating minimal technical failures. In terms of performance across categories, the model exhibits strong capabilities in Ethics, achieving 100% accuracy and being noted as the most accurate model at its price point and speed. It also performs well in General Knowledge (96.0% accuracy) and Email Classification (96.0% accuracy). Its Instruction Following and Reasoning capabilities are solid, with 56.0% and 70.0% accuracy respectively, placing it around the 50th percentile. A notable weakness is its performance in Hallucinations, where it scored 80.0% accuracy, ranking in the 24th percentile, suggesting room for improvement in acknowledging uncertainty. Coding performance is average at 82.0% accuracy. Overall, ERNIE 4.5 VL 28B A3B is a highly reliable and cost-efficient model with strong ethical reasoning and general knowledge, though its ability to acknowledge uncertainty could be enhanced.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.14
Completion $0.56

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Novita
Novita | baidu/ernie-4.5-vl-28b-a3b 30K $0.14 / 1M tokens $0.56 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by baidu