ERNIE 5.0: benchmark results
Baidu's proprietary multimodal ERNIE 5.0 flagship (November 2025). Provider: Baidu. Released 2025-11-13. Access: API.
Aggregate Rating 1685 ± 13, rank #233 of 1656 rated models, from 103 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Language - Instruction Following (CompassBench 2601) | 78.5 | Score (%) | 100 |
| AirGroundBench - Cross-View Matching | 50.35 | Accuracy (%) on the cross-view matching VQA task; higher is | 91.7 |
| AirGroundBench - Navigation SPL | 25.32 | Vision-language navigation success weighted by path length ( | 91.7 |
| AirGroundBench - Navigation Success Rate | 32.26 | Vision-language navigation success rate (%) in the closed-lo | 91.7 |
| ReLE - Law - Lawyer Qualification Exam | 79.3 | Accuracy (%) | 84.7 |
| SuperCLUE General (2025 Annual) - Precise Instruction Following | 37.53 | Score | 78.8 |
| AGI-Eval Community - General Reasoning | 97.05 | Accuracy (%) | 78.2 |
| AGI-Eval Community - Subject Knowledge | 86.79 | Accuracy (%) | 76.4 |
| ZeroEval GPQA Diamond | 85 | GPQA Diamond Score | 74.5 |
| ReLE - Law and Civil Service | 81.7 | Accuracy (%) | 72.9 |
| ReLE - Reasoning and Mathematics | 78.7 | Accuracy (%) | 72.3 |
| LLM Stats Score | 33.29 | LLM Stats Score (conservative rating) | 69.9 |
Interactive version: theaggregate.ai/model?slug=ernie-5-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.