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 · BenchGecko score 39.8. Top scores: OpenCompass — IFEval (81.2%), OpenCompass — AIME2025 (76.2%), OpenCompass — MMLU-Pro (70.8%).
Qwen3 30B A3B Instruct 2507 scores 40.1 (101% as good) at $0.05/1M input · 31% cheaper
OpenCompass Live Code Bench v6. Fresh competitive programming problems to evaluate code generation without memorization.
OpenCompass evaluation on AIME 2025 problems. Tests mathematical reasoning on fresh competition problems.
OpenCompass MMLU-Pro evaluation. Harder knowledge test with more answer choices.
OpenCompass evaluation of GPQA Diamond. PhD-level science questions from the hardest subset.
OpenCompass evaluation of Humanitys Last Exam. Expert-level cross-discipline knowledge test.
- Typetext
- Context131K tokens (~66 books)
- ReleasedOct 2025
- LicenseOpen Source
- StatusActive
- Cost / Message~$0.000
Frequently Asked Questions
Related Models
Claude 3.5 Sonnet · AnthropicGPT-4o (2024-08-06) · OpenAIQwen3.5-9B · Alibaba QwenQwen3 30B A3B Instruct 2507 · Alibaba QwenQwen3 14B · Alibaba QwenKey facts · as of 2026-05-03
- ERNIE 4.5 21B A3B Thinking by baidu. BenchGecko score 39.8, rank 199 of 312 scored models (normalized average of public benchmark scores).
- List price $0.0700 input · $0.28 output per 1M tokens (as of 2026-05-03).
How to cite · data as of 2026-05-03
ERNIE 4.5 21B A3B Thinking · benchmarks, pricing and providers. BenchGecko, data as of 2026-05-03. https://benchgecko.ai/model/ernie-4-5-21b-a3b-thinking
Credit "Source: BenchGecko" with a link. Prices per provider and Gecko Tests are BenchGecko data (CC BY 4.0); benchmark scores keep their original source, listed in the JSON. JSON · llms.txt · MCP