Yi (01.AI)
Freemium01.AI's open-weight LLM family founded by Kai-Fu Lee with strong bilingual Chinese/English performance
What is Yi (01.AI)?
01.AI is the AI startup founded by Kai-Fu Lee in 2023 that quickly became one of China's most notable LLM labs, releasing the Yi model family as open-weight models with strong bilingual Chinese and English capabilities. Yi models have shipped in multiple generations: Yi-6B, Yi-9B, Yi-34B (which briefly led the Hugging Face Open LLM Leaderboard), Yi-VL vision variants, and the Yi-Large flagship that powers 01.AI's commercial API. The Yi family is built on a Llama-compatible architecture, which means it plugs into the entire Llama tooling ecosystem, vLLM, Ollama, LM Studio, Hugging Face Transformers, and llama.cpp all support Yi out of the box with no special handling. 01.AI states in its official 01-ai/Yi repository that the code and weights of the Yi series are distributed under the Apache 2.0 license, so personal, academic, and commercial use are all covered. Where Yi really shines is bilingual performance: it is one of the strongest openly-available models for Chinese language tasks while maintaining competitive English capabilities, making it a go-to choice for applications targeting Chinese-speaking users, translation, and multilingual RAG systems. Status note, checked August 2026: 01.AI has pivoted away from building its own frontier models. The company stopped pre-training large language models in 2025 and now sells enterprise decision products (Boss AI, TrueNorth, TopSales AI, WorldWise) built largely on other labs' open models, its 01.ai site leads with that enterprise suite rather than with Yi, the newest Yi checkpoints on Hugging Face still date to 2024, and the global hosted API is gone: platform.01.ai says its API services are unavailable and directs users to 01.AI's Chinese platform, while api.01.ai no longer resolves. The open weights themselves remain published and downloadable, so Yi is still a reasonable self-hosted pick for bilingual Chinese and English work alongside Qwen and DeepSeek, both of which are shipping new models far more often.
⚡ Quick Verdict
Bilingual Chinese/English applications and teams needing strong open models with Llama-compatible architecture
English-only users who want the biggest fine-tune ecosystem
Free open weights under Apache 2.0 · global hosted API offline as of August 2026
Yes, open weights are free
Best-in-class bilingual Chinese/English performance in an open-weight LLM
Development has stalled, no new Yi base models since 2024, and a smaller English community than Llama or Mistral
Bottom line: Yi scores 4.3/5, still a capable self-hosted open-weight family for bilingual Chinese and English work, but a frozen one: 01.AI stopped pre-training its own models in 2025 and the global hosted API is gone. Choose Yi-34B for general use, Yi-34B-200K for long documents, and check newer open models before committing.
Pricing
Open weights, Free: Yi-6B, Yi-9B, Yi-34B, the Yi-1.5 refresh, the 200K long-context builds, and Yi-VL can be downloaded from the 01-ai organization on Hugging Face or GitHub. 01.AI distributes the Yi code and weights under the Apache 2.0 license, which covers personal, academic, and commercial use. The most recent checkpoint updates in that organization date to 2024.
01.AI hosted API, discontinued globally: platform.01.ai now shows a notice that its API services are unavailable and invites users to register on 01.AI's Chinese platform at platform.lingyiwanwu.com instead. The former global endpoint api.01.ai does not resolve as of August 2026. Our May 2026 figure of roughly $0.21 per million tokens for Yi Base 6B came from third-party API marketplaces rather than 01.AI, and it could not be re-verified against any 01.AI price list in August 2026, so treat it as historical.
Third-party inference: Third-party hosts such as Together AI, Fireworks, DeepInfra, and OpenRouter have carried Yi-34B and smaller variants at per-token rates; check each provider's current catalogue, since availability of older models moves around. Self-hosting is free beyond hardware costs.
Key Features
- Yi-6B, Yi-9B, Yi-34B open checkpoints
- Yi-VL vision variants with multimodal understanding
- Yi-Large flagship model, API access now only via 01.AI's Chinese platform
- Llama-compatible architecture, works with all Llama tooling
- Class-leading bilingual Chinese/English performance
- 200K context window support in Yi-34B-200K variant
- Backed by Kai-Fu Lee and Sinovation Ventures
- Apache 2.0 licensed code and weights
Pros & Cons
Pros
- Best-in-class open model for Chinese language tasks
- Llama-compatible, instant ecosystem support
- 200K context window in the 34B-200K variant
- Apache 2.0 weights, free for commercial use, with strong English performance too
Cons
- Smaller English-language ecosystem than Llama
- Fewer fine-tunes and community derivatives
- Development has stalled, no new Yi base models since 2024 and 01.AI stopped pre-training its own LLMs in 2025
- Global hosted API is gone, API users are sent to 01.AI's Chinese platform
FAQ
Who is behind 01.AI?
01.AI was founded in 2023 by Kai-Fu Lee, a well-known AI researcher who previously led Google China and Microsoft Research Asia. The company is backed by Sinovation Ventures and reached unicorn status within nine months of founding, with a mission to build globally competitive open-weight LLMs from China. The Yi model family is 01.AI's flagship output.
Is Yi better for Chinese than Qwen?
Both Yi and Qwen are excellent for Chinese language tasks, and the winner depends on the specific benchmark. Qwen currently has more frequent updates and a larger ecosystem, while Yi is sometimes preferred for bilingual Chinese-English tasks where maintaining English quality matters. For most production Chinese deployments, try both on your specific workload and pick the one that scores higher.
Does Yi work with Ollama and LM Studio?
Yes, smoothly. Yi uses a Llama-compatible architecture, so any tool that loads Llama models, Ollama, LM Studio, llama.cpp, vLLM, Hugging Face Transformers, supports Yi out of the box. Just use the 01-ai/Yi-34B or similar model ID.
What is Yi-34B-200K?
Yi-34B-200K is a long-context variant of Yi-34B that supports up to 200,000 tokens of input context. It is designed for document analysis, long-form RAG, and any workload that benefits from very long prompts. At release it was one of the first openly available 200K-context models, and it remains competitive for open-weight long-context use cases in 2026.
How does Yi-Large compare to GPT-4?
Yi-Large is 01.AI's frontier commercial model, launched in 2024. It scored competitively with the GPT-4 of that era on several Chinese benchmarks and performed respectably on English ones, but it has not been refreshed since, so it now trails the current frontier models from OpenAI, Anthropic, and Google. Access has also narrowed: as of August 2026 platform.01.ai says its API services are unavailable and directs users to 01.AI's Chinese platform at platform.lingyiwanwu.com, and the old global endpoint api.01.ai no longer resolves.
Can I fine-tune Yi commercially?
Yes. 01.AI states in the official 01-ai/Yi repository that the code and weights of the Yi series models are distributed under the Apache 2.0 license, so personal, academic, and commercial use are all permitted, including commercial fine-tuning and redistribution of fine-tuned weights. Some earlier 2023 checkpoints shipped under the separate Yi Series Models Community License, so check the license file bundled with the exact checkpoint you download.
📋 Good to know
Download from Hugging Face (01-ai organization), load with standard Transformers or Ollama, no special configuration needed.
Self-hosting keeps everything local, which is now the practical route: 01.AI's global API platform is offline and remaining hosted access runs through the company's Chinese platform, which carries Chinese data handling rules.
Move from Yi-6B to Yi-34B for stronger reasoning. For frontier-class tasks, note that Yi-Large is a 2024 model and its API is no longer offered globally, so a current model from another lab is usually the better upgrade.
Very low, Llama-compatible means any Llama tutorial works for Yi.
