Benchmark Almanac

ERNIE 5.0

ERNIE 5.0 ranks first on 3 of the 41 benchmarks with five or more models scored.

Baidu Released January 22nd, 2026 Family ERNIE 5 The lab's model card Compare with others
48
Benchmarks with a score
3
Ranked first, out of five or more models
9
In the top three
Baidu Jan 2026

ERNIE 5.0

Closed weights Flagship

145.8 #124 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • MMMU-Pro 68.6% #54 /90
  • BFCL V4 66.5% #13 /30
  • SimpleVQA 67.6% #11 /30

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
Multi-IF 85.6% #2 of 28 Lab model card
Chinese SimpleQA 86% #1 of 13 Lab model card
IFEval 93.3% #15 of 135 Lab model card
SimpleQA 74% #8 of 60 Lab model card
BBEH micro average 66.6% #2 of 15 Lab model card
MultiChallenge 66% #10 of 42 Lab model card
MMLU-Pro 83.8% #45 of 167 Lab model card
SimpleVQA 67.6% #11 of 30 Lab model card
HallusionBench 63.9% #17 of 31 Lab model card
Humanity's Last Exam 25.8% #89 of 154 Lab model card
Extended NYT Connections 10.3% #98 of 103 Official leaderboard

Math and science

BenchmarkScoreRankSource
ZebraLogic 96.5% #2 of 23 Lab model card
GPQA Diamond 86.4% #66 of 302 Lab model card
AIME 2025 89.1% #56 of 143 Lab model card
HMMT 2025 79.6% #37 of 54 Lab model card

Coding

BenchmarkScoreRankSource
HumanEval+ 94.5% #1 of 27 Lab model card
MBPP+ 82.5% #2 of 15 Lab model card
LiveCodeBench v6 76.2% #41 of 90 Lab model card

Agents and tools

BenchmarkScoreRankSource
AceBench 88.7% #1 of 3 Lab model card
BFCL V4 66.5% #13 of 30 Lab model card
tau2-bench 78.8% #43 of 79 Lab model card
BrowseComp-ZH 64.7% #11 of 19 Lab model card

Professional work

BenchmarkScoreRankSource
SpreadsheetBench 40.1% #6 of 6 Lab model card

Vision, audio and robotics

BenchmarkScoreRankSource
AI2D 96.9% #1 of 52 Lab model card
MathVerse 85.1% #2 of 12 Lab model card
MMAU 80.4% #3 of 15 Lab model card
DocVQA 95.5% #8 of 36 Lab model card
BLINK 70% #7 of 29 Lab model card
AISHELL-1 0.31% #1 of 4 Lab model card
CVBench 87.2% #4 of 15 Lab model card
MathVista 84.8% #19 of 70 Lab model card
CountBench 96.5% #8 of 28 Lab model card
Clotho-AQA 73.7% #1 of 3 Lab model card
GenEval 90.1% #1 of 3 Lab model card
VisualPuzzles 64.8% #1 of 3 Lab model card
ChartQA 87.8% #13 of 37 Lab model card
OCRBench 87.8% #12 of 34 Lab model card
MMVU 72.2% #12 of 26 Lab model card
MMStar 75.5% #19 of 36 Lab model card
Video-MME 81.3% #27 of 51 Lab model card
MATH-Vision (MathVision) 74.3% #28 of 51 Lab model card
Video-MMMU 81.1% #21 of 36 Lab model card
MMMU-Pro 68.6% #54 of 90 Lab model card
AISHELL-2 2.64% #2 of 3 Lab model card
VLMs Are Blind 91.4% #10 of 14 Lab model card
VisuLogic 32% #5 of 7 Lab model card
CharXiv 67.1% #58 of 75 Lab model card
VBench 84.2% #1 of 1 Lab model card

Its lab's numbers put it at 145.8, number 124 of 404

The Self-Reported Index fits one score to the 38 capability results Baidu reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 144.0 to 147.5. How we compute it.

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

ERNIE 5.0: A 2.4 Trillion-Parameter Unified Multimodal Foundation Model (ERNIE Blog) Feb 6, 2026

Its system card

Every result in the longer documents the lab published with it. 9 of them aren't in the launch post.

ERNIE 5.0 Technical Report Feb 4, 2026

Capabilities