01.AI open-sources Yi-6B and Yi-34B
Kai-Fu Lee's 01.AI released its first open-weight models, which it said outperformed larger Llama 2 and Falcon models.
- Open weights & ecosystem
- Models & capabilities
- Minor
01.AI, the Beijing-based startup founded seven months earlier by former Google China head Kai-Fu Lee, released its first models: Yi-6B and Yi-34B, open-weight bilingual (English/Chinese) base models under the Apache 2.0 licence. The models used a Llama-style architecture but were trained from scratch rather than derived from Meta’s weights, on a corpus the company described as 3 trillion tokens.
01.AI published benchmark comparisons showing Yi-34B ahead of Meta’s Llama 2-70B and TII’s Falcon-180B on several standard evaluations despite having a fraction of their parameter counts, and the base model topped the Hugging Face Open LLM Leaderboard rankings for open models at release. Lee framed the decision to debut with an open release, rather than a closed product, as a way for the company to “give back” to the field it was entering late.
The release came alongside reports that 01.AI had reached a valuation of roughly $1 billion within months of founding, making it one of several Chinese labs — alongside DeepSeek and others — establishing open-weight releases as a competitive strategy against better-funded, closed US labs. Extended-context variants (Yi-34B-200K and Yi-6B-200K) followed, widening the gap with contemporaries that had not yet shipped comparable long-context open models.