Baidu: ERNIE 4.5 300B A47B

Baidu: ERNIE 4.5 300B A47B

baidu · Released Jun 30, 2025
25
Our Score

ERNIE-4.5-300B-A47B is a 300B parameter Mixture-of-Experts (MoE) language model developed by Baidu as part of the ERNIE 4.5 series. It activates 47B parameters per token and supports text generation in both English and Chinese. Optimized for high-throughput inference and efficient scaling, it uses a heterogeneous MoE structure with advanced routing and quantization strategies, including FP8 and 2-bit formats. This version is fine-tuned for language-only tasks and supports reasoning, tool parameters, and extended context lengths up to 131k tokens. Suitable for general-purpose LLM applications with high reasoning and throughput demands.

$0.28 / 1M Input Price
$1.10 / 1M Output Price
123,000 tokens Context Window
12,000 tokens Max Output
300B Parameters

Architecture

ModalityText → Text
TokenizerOther
Parameters300B

Performance Indices

Source: Artificial Analysis

15 Intelligence Index
14.5 Coding Index
6.1 Agentic Index
41.3 Math Index

Benchmark Scores

Evaluations

GPQA Diamond 81.1%
Graduate-level scientific reasoning
HLE 3.5%
Humanity's Last Exam
MMLU Pro 77.6%
Multi-task language understanding
LiveCodeBench 46.7%
Live coding evaluation
SciCode 31.5%
Scientific computing
MATH 500 93.1%
Mathematical problem-solving
AIME 49.3%
Competition mathematics
AIME 2025 41.3%
Competition mathematics (2025)
IFBench 39.1%
Instruction following
LCR 2.3%
Long-context reasoning
TerminalBench Hard 6.1%
Agentic terminal tasks

Benchmark data from Artificial Analysis and Hugging Face

Model Information

OpenRouter ID baidu/ernie-4.5-300b-a47b
Providerbaidu
Release Date June 30, 2025
Context Length123,000 tokens
Max Completion12,000 tokens
Status Active

Pricing

Token Type Cost per 1M tokens Cost per 1K tokens
Input $0.28 $0.000280
Output $1.10 $0.001100

Live Performance

Live endpoint metrics — refreshed every 30 minutes.

1,189ms
Best Latency (TTFT)
32 tok/s
Best Throughput
0/2
Active Endpoints
Available via: Novita, SiliconFlow