How much does ERNIE 4.5 21B A3B Thinking cost?

ERNIE 4.5 21B A3B Thinking costs $0.07 per million input tokens and $0.28 per million output tokens. For a typical conversation (~2,000 tokens), that's approximately $0.000 per message.

What benchmarks has ERNIE 4.5 21B A3B Thinking been tested on?

ERNIE 4.5 21B A3B Thinking has been evaluated on 6 benchmarks. Top scores: OpenCompass — IFEval: 81.2, OpenCompass — AIME2025: 76.2, OpenCompass — MMLU-Pro: 70.8.

Is ERNIE 4.5 21B A3B Thinking open source?

Yes, ERNIE 4.5 21B A3B Thinking is open source.

How does ERNIE 4.5 21B A3B Thinking compare to Grok 3?

ERNIE 4.5 21B A3B Thinking has an average score of 39.8 while Grok 3 scores 39.9. Grok 3 slightly outperforms ERNIE 4.5 21B A3B Thinking overall. ERNIE 4.5 21B A3B Thinking costs $0.07/1M input vs Grok 3 at $3.00/1M input. See full comparison →

Home/Models/ERNIE 4.5 21B A3B Thinking

ERNIE 4.5 21B A3B Thinking

Name: ERNIE 4.5 21B A3B Thinking
Price: 0.07 USD
Author: baidu

by baidu · Released Oct 2025

Open Source

39.8

avg score

Rank #146

Compare

Better than 37% of all models

Context

131K tokens (~66 books)

Input $/1M

$0.07

Output $/1M

$0.28

Type

text

License

Open Source

Benchmarks

6 tested

Data updated today

About

ERNIE-4.5-21B-A3B-Thinking is Baidu's upgraded lightweight MoE model, refined to boost reasoning depth and quality for top-tier performance in logical puzzles, math, science, coding, text generation, and expert-level academic benchmarks.

Tested on 6 benchmarks with 57.0% average. Top scores: OpenCompass — IFEval (81.2%), OpenCompass — AIME2025 (76.2%), OpenCompass — MMLU-Pro (70.8%).

Looking for similar performance at lower cost?
Qwen3 8B scores 40.6 (102% as good) at $0.05/1M input · 29% cheaper

Capabilities

coding

47.0

#76 globally

math

76.2

#24 globally

knowledge

45.8

#122 globally

language

81.2

#48 globally

Benchmark Scores

Compare All

Tested on 6 benchmarks · Ranked across 4 categories

Score Distribution (all 233 models)

0255075100

▲ You are here

codingCompare coding →

OpenCompass — LiveCodeBenchV6

OpenCompass Live Code Bench v6. Fresh competitive programming problems to evaluate code generation without memorization.

47.0—

mathCompare math →

OpenCompass — AIME2025

OpenCompass evaluation on AIME 2025 problems. Tests mathematical reasoning on fresh competition problems.

76.2—

knowledgeCompare knowledge →

OpenCompass — MMLU-Pro

OpenCompass MMLU-Pro evaluation. Harder knowledge test with more answer choices.

70.8—

OpenCompass — GPQA-Diamond

OpenCompass evaluation of GPQA Diamond. PhD-level science questions from the hardest subset.

60.2—

OpenCompass — HLE

OpenCompass evaluation of Humanitys Last Exam. Expert-level cross-discipline knowledge test.

6.5—

Quick compare:

vs Grok 3

vs GPT-4o (2024-08-06)

vs Qwen3 30B A3B Instruct 2507

Excellent (85+) Good (70-85) Average (50-70) Below (<50)

Similar Models

Qwen3 30B A3B Instruct 2507

Alibaba Qwen

40.1$0.09/1M

Links

Info

baidu Pricing explorer Developers · API

Research

Documentation

Community

Source Code

BenchGecko API

ernie-4-5-21b-a3b-thinking

Specifications

Typetext
Context131K tokens (~66 books)
ReleasedOct 2025
LicenseOpen Source
StatusActive
Cost / Message~$0.000

Available On

baidu$0.07

Frequently Asked Questions

ERNIE 4.5 21B A3B Thinking is an open-source text AI model by baidu, released in October 2025. It has an average benchmark score of 39.8. Context window: 131K tokens.

Benchmarks

OpenCompass — IFEval OpenCompass — AIME2025 OpenCompass — MMLU-Pro OpenCompass — GPQA-Diamond OpenCompass — LiveCodeBenchV6

baidu · Provider baidu · Economy All Models Compare Models Pricing Developers · API Context Window · Glossary

ERNIE 4.5 21B A3B Thinking

Frequently Asked Questions

Related Models

Benchmarks

Related Pages