ERNIE 5.0

ERNIE 5.0 is Baidu's flagship model, unveiled at Baidu World 2025. It is a natively omni-modal system with up to 2.4 trillion parameters that jointly models text, images, audio, and video in one architecture rather than bolting modalities onto a text backbone. Baidu positions it against GPT-5-class and Gemini-class models, citing 40-plus benchmark tests where its language and multimodal understanding sit near the top, and image/video generation it calls world-leading in professional use. A January 2026 snapshot topped the LMSYS Arena text leaderboard at 1,451 points, per third-party coverage. It is proprietary: access runs through the ERNIE app for consumers and Baidu's Qianfan cloud platform for developers, with reported pricing around $0.85 per million input tokens. Baidu leans on its knowledge graph and Chinese-language data as differentiators. The announcement was met with some market skepticism about differentiation, and independent verification of the headline benchmark claims remains thin.

Vendor-stated: competitive with Gemini 2.5 Pro and GPT-5-High across 40+ tests; 1,451 LMSYS Arena text (third-party reported, Jan 2026)