Yi-1.5 34B

Yi-1.5 34B is 01.AI’s updated 34B line. Its model card describes continued pretraining on 500 billion high-quality tokens and fine-tuning on 3 million examples. 01.AI says the process improves coding, mathematics, reasoning, and instruction following while preserving language understanding and common sense. Its context window is 4,096 tokens.

Compared with the older Yi-34B, Yi-1.5 34B is explicitly documented as an updated training stage. The Yi-1.5 family still has separate base and chat variants.