01.AI Yi

Yi-34B and the Yi-1.5 refresh were genuinely strong for their release window: Yi-34B topped the Hugging Face Open LLM Leaderboard for pretrained base models in November 2023, and Yi-34B-Chat placed second on the AlpacaEval leaderboard behind GPT-4 Turbo (data as of January 2024), per Hugging Face's own tracking (https://huggingface.co/01-ai). The whole line ships under Apache-2.0 with no usage restrictions, which is more permissive than Meta's Llama license, and 01.AI's OpenAI-compatible API (api.lingyiwanwu.com) lets you swap in Yi with a couple of lines of SDK config (https://platform.lingyiwanwu.com/docs). For a buyer who wants a bilingual (Chinese/English) open-weight model with no license friction and doesn't need current-generation performance, that combination is still real.

The main GitHub repository (01-ai/Yi, 7,836 stars, Apache-2.0) has not been pushed to since 27 November 2024, per GitHub's API (https://api.github.com/repos/01-ai/Yi). It is not archived — issues and pull requests can still be opened — but nothing has been merged since that date either: the most recent closed pull requests (docs fixes submitted in July and August 2026) were closed without merging (https://github.com/01-ai/Yi/pull/653, https://github.com/01-ai/Yi/pull/649). The Hugging Face model cards show the same pattern: the last Yi-series weights (Yi-34B-Chat, Yi-9B, Yi-6B and variants) were updated 11 November 2024, and Yi-1.5 stopped at 26 June 2024 (https://huggingface.co/01-ai/models). There is no GitHub Releases history for the repo at all. None of that makes the existing weights unusable — Apache-2.0 code that still runs is still Apache-2.0 code — but a buyer evaluating Yi today should not expect bug fixes, new checkpoints, or support for current inference tooling versions.

01.AI's own homepage (01.ai) no longer markets Yi or API access to it at all — the entire product lineup is now enterprise decision-support agents (Boss AI, Investor AI, TopSales AI, under a platform called TrueNorth, released July 2026) and government/industry AI transformation consulting, not foundation models (https://www.01.ai/). The Financial Times reported in March 2025 that 01.AI stopped pretraining large language models altogether and pivoted to selling tailored AI business solutions built on DeepSeek's models instead (https://www.ft.com/content/c19f3988-45d7-4a81-854d-9ba0d71812fe, via Wikipedia summary at https://en.wikipedia.org/wiki/01.AI). That pivot lines up with what's on the site today: a reader looking for "01.AI" now finds a sovereign-AI and enterprise-agent consultancy, with Yi as a legacy open-source artifact rather than the product being sold.

01.AI's own current API documentation states that requests to the yi-lightning model are intelligently routed to DeepSeek-V3, Qwen3-30B-A3B, or Yi-Lightning depending on the input, to give a better cost/performance trade-off; the vision model yi-vision-v2 is routed similarly, to Qwen2.5-VL-72B-Instruct or Yi-Vision-V2 (https://platform.lingyiwanwu.com/docs, fetched 2026-09-12). In plain terms: calling 01.AI's flagship API model today does not guarantee you get 01.AI's own weights back — you may get a different vendor's open model, chosen by 01.AI's router. Pricing is listed plainly on the same page: ¥0.99 per million tokens for yi-lightning, ¥6 per million tokens for yi-vision-v2.

On LMArena's live text leaderboard, yi-lightning — the model 01.AI announced in October 2024 as ranking sixth globally on the same leaderboard — currently places 232nd out of roughly 340 ranked models; yi-1.5-34b-chat ranks 327th and yi-34b-chat ranks 340th (https://lmarena.ai/leaderboard/text, checked 2026-09-12). This isn't a knock on the models' 2024 results, which were real; it reflects how fast the leaderboard has filled in with newer releases from every major lab since. A buyer comparing Yi to what's available today, rather than to 2024's field, should use this current standing rather than the vendor's original launch-day framing.

01.AI was founded in Beijing in March 2023 by Kai-Fu Lee (former Google China president and Microsoft Research Asia founder) and reached a valuation over $1 billion within eight months, per Bloomberg (https://www.bloomberg.com/news/articles/2023-11-05/kai-fu-lee-s-open-source-01-ai-bests-llama-2-according-to-hugging-face, via https://en.wikipedia.org/wiki/01.AI). External backers reported include Alibaba Group and Xiaomi (Wall Street Journal, via the same Wikipedia summary). 01.AI is grouped among China's so-called "Six AI Tigers" alongside Zhipu AI, Moonshot AI, MiniMax, Baichuan, and StepFun. Not established: a current post-2023 valuation or total funding raised — no primary source with an up-to-date figure was found.