Baidu's Ernie 5.1 reportedly cuts pre-training costs 94% via 'Once-For-All' technique

Baidu's Ernie 5.1 uses one third of its predecessor's parameters and reportedly cost just 6% of comparable models to pre-train, thanks to a 'Once-For-All' training method that extracts multiple sub-models from a single training run. It ranks 4th on Search Arena behind two Claude Opus variants and GPT-5.5 Search.