← Back to Discover

ERNIE

Official ERNIE site from Baidu, a blog announcing the ERNIE 5.x models, the ERNIE-Image text-to-image model and the PaddleOCR-VL-1.5 document parsing model.

General assistants
Visit website ↗

What it does

ERNIE is a model family from Baidu, and this site is its blog. The latest post announces ERNIE 5.1, released on 9 May 2026, which the blog says reaches leading results at only 6% of the pre-training cost of comparable models and improves agent, reasoning and creative abilities. Earlier posts introduce ERNIE 5.0, a 2.4 trillion-parameter model trained from scratch that handles text, image, video and audio in one autoregressive framework; ERNIE-Image, a text-to-image model with 8B DiT parameters that the blog says leads among open-weights models; and PaddleOCR-VL-1.5, a 0.9B vision-language model for parsing documents that scores 94.5% on OmniDocBench v1.5. Several short posts report ERNIE preview and release versions placing first among Chinese models on LMArena leaderboards. The page does not describe pricing, apps or API access, so how to use each model is not verified from this page.

Capabilities

  • ERNIE 5.1 announced as officially released on 9 May 2026
  • ERNIE 5.0 is a 2.4 trillion-parameter model covering text, image, video and audio
  • ERNIE-Image is an 8B-parameter text-to-image model described as leading among open-weights models
  • PaddleOCR-VL-1.5 is a 0.9B model for document parsing
  • Posts track ERNIE rankings on the LMArena text and vision leaderboards

When to use it

  • You want to follow which ERNIE models Baidu has released and how they are benchmarked
  • You are looking for a text-to-image model described as leading among open-weights models to evaluate
  • You need a small vision-language model for parsing documents