Benchmark Almanac

Baidu

6 models 6 model cards read ernie.baidu.com/blog/

What its model cards measure

Tall ticks are Baidu model cards that list the benchmark in their headline table. Gray dots are cards that leave it out. Rows are the benchmarks the lab listed most, at least twice each, in the order it started using them.

MMLU-Pro Chinese SimpleQA HumanEval+ OCRBench ChartQA MathVista DocVQA GPQA Diamond BIG-Bench Hard MBPP+ MMMU C-Eval CMMLU MATH-500 GSM8K CMATH DROP LiveCodeBench IFEval Multi-IF AIME 2025 ZebraLogic SimpleQA MuSR MathVerse MATH-Vision (MathVision) VLMs Are Blind VisuLogic

Model cards

DateModelDocumentBenchmarks
9 May 2026 ERNIE 5.1 ERNIE 5.1 Officially Released! Topping Multiple Leaderboards (ERNIE Blog) 8
6 Feb 2026 ERNIE 5.0 ERNIE 5.0: A 2.4 Trillion-Parameter Unified Multimodal Foundation Model (ERNIE Blog) 43
11 Nov 2025 ERNIE-4.5-VL-28B-A3B-Thinking Introducing ERNIE-4.5-VL-28B-A3B-Thinking (Hugging Face model card and ERNIE Blog) 19
9 Sep 2025 ERNIE-4.5-21B-A3B-Thinking ERNIE-4.5-21B-A3B-Thinking model card (Hugging Face) 11
30 Jun 2025 ERNIE-4.5-300B-A47B Announcing the Open Source Release of the ERNIE 4.5 Model Family (ERNIE Blog) 24
16 Mar 2025 ERNIE 4.5 两连发！文心大模型4.5及X1，上线千帆！ (Baidu AI Cloud news) 22

Every Baidu model