ERNIE 5.1

ERNIE 5.1 (May 2026) introduces an elastic training framework with parameter compression to one-third of prior scale and training cost at roughly 6% of peers, while strengthening creative writing, deep search, and agent tasks.

It targets content teams, research assistants, and agent workflows that need reliable retrieval, generation, and multi-step execution under controlled training economics.

ERNIE 5.1 demonstrates that frontier-style writing, search, and agent capability can be pursued at substantially lower training cost, supporting the view that relatively low-cost models can still achieve high performance.