01.AI · General LLM
Yi
A bilingual Chinese-English language model family from Kai-Fu Lee's 01.AI, known for strong benchmark performance across diverse tasks.
Overview
Yi is a family of large language models developed by 01.AI, founded by AI pioneer Kai-Fu Lee. The Yi model family spans from 6B to 34B parameters and demonstrates impressive performance on both Chinese and English benchmarks, often outperforming larger models. Yi-1.5 introduced improved reasoning and instruction following, while Yi-VL added vision-language capabilities. The models are designed for both research and commercial use with a permissive Apache 2.0 license.
Parameters
6B / 9B / 34B variants
Context Window
200K tokens (Yi-1.5)
Architecture
Dense decoder-only transformer
Training Data
3.1T tokens (English, Chinese, code)
License
Apache 2.0
Capabilities
Strong bilingual Chinese-English text generation
Competitive reasoning and knowledge benchmarks
Vision-language understanding (Yi-VL)
Instruction following and conversational AI
Use Cases
Building bilingual AI applications for Chinese and English markets
Deploying efficient language models for general-purpose tasks
Creating vision-language applications with Yi-VL
Fine-tuning for domain-specific Chinese or English applications
Pros
- +Strong bilingual performance with Apache 2.0 license
- +Excellent benchmark results relative to model size
- +200K context window in Yi-1.5 for long-document tasks
- +Active development with regular model improvements
Cons
- -Less community adoption than Llama or Mistral models
- -Limited enterprise support infrastructure outside of China
- -Vision-language variant is less mature than text-only models
- -Brand recognition lower in Western markets
Pricing
Free and open-source under Apache 2.0. Self-hosting costs vary. Available through various cloud inference providers at competitive per-token rates.