Baidu: ERNIE 4.5 21B A3B

Text input Text output
Author's Description

A sophisticated text-based Mixture-of-Experts (MoE) model featuring 21B total parameters with 3B activated per token, delivering exceptional multimodal understanding and generation through heterogeneous MoE structures and modality-isolated routing. Supporting an extensive 131K token context length, the model achieves efficient inference via multi-expert parallel collaboration and quantization, while advanced post-training techniques including SFT, DPO, and UPO ensure optimized performance across diverse applications with specialized routing and balancing losses for superior task handling.

Key Specifications
Cost
$$
Context
120K
Parameters
21B
Released
Aug 12, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Stop Presence Penalty Logit Bias Top P Temperature Seed Min P Frequency Penalty Max Tokens
Performance Summary

Baidu's ERNIE 4.5 21B A3B model demonstrates competitive response times, ranking in the 48th percentile for speed across various benchmarks. It consistently offers highly competitive pricing, placing in the 82nd percentile. The model exhibits exceptional reliability with a 97% success rate, indicating minimal technical failures and consistent provision of usable responses. Performance across benchmark categories reveals a mixed profile. While excelling in Instruction Following (54th percentile accuracy) and demonstrating solid General Knowledge (43rd percentile accuracy), its performance in Coding (44th percentile accuracy) and Reasoning (24th percentile accuracy) is less robust. Notably, the model shows lower accuracy in Classification (Email Classification: 28th percentile) and Ethics (28th percentile), despite high raw scores, suggesting these areas are highly competitive. Its key strengths lie in its cost-effectiveness and high reliability, making it a dependable choice for applications where consistent operation and budget are critical. Weaknesses are primarily in complex reasoning and specific classification tasks where it lags behind top performers.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.07
Completion $0.28

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Novita
Novita | baidu/ernie-4.5-21b-a3b 120K $0.07 / 1M tokens $0.28 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration
Other Models by baidu