Definition
Yi is a family of large language models developed by 01.AI, a Chinese AI company founded by Kai-Fu Lee, offering open-weight bilingual Chinese-English models (such as Yi-6B and Yi-34B) alongside a larger proprietary model, Yi-Large, accessible through 01.AI's commercial API.
Overview
Yi was launched by 01.AI in 2023, one of a wave of well-funded Chinese AI startups founded in the wake of ChatGPT's release. 01.AI was founded by Kai-Fu Lee, a prominent AI investor and former head of Google China and Microsoft Research Asia, positioning Yi as a serious commercial and research effort from its inception rather than a purely academic project. The initial Yi release included open-weight base and chat models at 6B and 34B parameter scales, trained on a large bilingual Chinese-English corpus and licensed for both research and, with registration, commercial use. At release, Yi-34B drew attention for achieving competitive results against larger contemporary open models on English and Chinese benchmarks, credited in part to careful data curation and pretraining recipe choices similar in spirit to approaches used by Meta's Llama team, several of whose techniques 01.AI's team studied closely. 01.AI subsequently expanded the line with Yi-VL (vision-language multimodal variants), longer-context versions, and Yi-Large, a larger, more capable proprietary model offered exclusively through 01.AI's API rather than as open weights, aimed at competing with flagship models from both Chinese and Western labs on general capability benchmarks. Yi models are notable for their explicit bilingual design goal — performing strongly in both Chinese and English rather than treating one as a secondary language — which distinguishes them from many Western-developed open models that handle non-English languages as a secondary training objective. This makes Yi models a popular choice for developers building products serving both Chinese and English-speaking users from a single model.
Key Features
- Developed by 01.AI, founded by AI researcher and investor Kai-Fu Lee
- Open-weight models available at multiple scales, including Yi-6B and Yi-34B
- Strong, explicitly designed bilingual Chinese-English performance
- Yi-Large proprietary flagship model offered via commercial API
- Yi-VL multimodal vision-language variants
- Competitive benchmark performance against larger contemporary open models at release
- Data curation and training recipe influenced by best practices from leading open-model efforts
- Licensed for both research and, with registration, commercial use for open variants