01.AI

01.AI is a Chinese artificial intelligence company headquartered in Beijing and founded in March 2023 by Kai-Fu Lee, the former Microsoft and Google executive and co-founder of Sinovation Ventures. Its Chinese name, 零一万物, is commonly rendered as Lingyi Wanwu and conveys the idea of “zero-one, everything,” an allusion to Taoist philosophical language. The company emerged during a period of intense competition among Chinese AI start-ups and is commonly grouped with the country’s leading “AI Tigers.” The company’s initial strategy centered on open-source large language models. Rather than concentrating exclusively on the largest and most expensive systems, 01.AI argued that many developers and businesses need capable, comparatively efficient models that can be deployed at lower cost. This approach led to the Yi family of models, including bilingual Chinese-English systems and specialized coding models. The company also developed Wanzhi, a productivity assistant offered in Chinese and English and positioned as a general-purpose workplace tool. 01.AI’s early expansion was shaped by restricted access to advanced semiconductor hardware. In response to United States export controls and limited chip availability, the company reportedly accumulated processors, borrowed funding from Sinovation Ventures to support purchases, and emphasized more efficient infrastructure and inference engines. The company’s engineering approach included efforts to reduce hardware waste and maintain a lower-than-industry-average chip-cluster failure rate. These constraints became part of its broader emphasis on efficiency rather than simply scaling compute. The company began operations in June 2023. By November of that year, it had reportedly reached a valuation above one billion US dollars within eight months of launch, giving it unicorn status, and had grown to more than 100 employees. External backers reported in reference material include Alibaba Group and Xiaomi, although the available material does not establish their precise ownership stakes or governance roles. 01.AI’s product development continued through 2024. Yi-34B, released in November 2023, was made available globally in Chinese and English and was reported to perform strongly against other open-source models, including Meta’s Llama 2, on selected benchmarks. Yi-Coder followed in September 2024 as a coding model supporting 52 programming languages, with 1.5-billion- and 9-billion-parameter versions and a 128,000-token context window. Yi-Lightning, released in October 2024, targeted high-performance, lower-cost inference and received a strong placement in a ranking associated with UC Berkeley SkyLab and LMSYS. During late 2024 and early 2025, reporting portrayed the company as increasingly focused on cost control and commercial application. It was described as using smaller training datasets and lower-cost engineering resources, while a November 2024 report said its models had been trained with approximately 2,000 GPUs at a cost substantially below that associated with some leading United States model developers. In January 2025, Kai-Fu Lee rejected reports that the company was negotiating to sell its pre-training team to Alibaba Cloud. In March 2025, 01.AI reportedly stopped pre-training large language models as an independent priority and shifted toward selling customized business solutions built using DeepSeek models. This represented a move from primarily developing foundation models toward applied enterprise AI services.

01.AI was established in Beijing in March 2023 by Kai-Fu Lee, a technology executive and venture investor whose previous career included leadership roles at Microsoft and Google and the co-founding of Sinovation Ventures. The company’s Chinese name, 零一万物, refers to the relationship between “zero-one” and “everything,” with a stated connection to Taoist thought. From its beginning, 01.AI positioned itself within China’s rapidly developing generative-AI sector and became associated with the group of leading Chinese AI start-ups sometimes called the “AI Tigers.” The company started operations in June 2023. Its first major strategic choice was to focus on open-source language models. This approach reflected the view that most developers and many businesses do not require the biggest available model and instead benefit from systems that are capable, affordable, and easier to adapt or deploy. 01.AI nevertheless indicated that proprietary models could be developed for customers in the future, leaving room for a commercial enterprise-model business alongside its open-source work. Hardware access was a major early constraint. United States controls on advanced semiconductors limited the supply of chips available to Chinese AI companies. Reference material describes 01.AI as purchasing and stockpiling semiconductors earlier in 2023 and borrowing from Sinovation Ventures to help finance those purchases. Because its chip supply was limited, the company concentrated on infrastructure efficiency and inference-engine design. It also reportedly maintained a chip-cluster failure rate below the industry average, making operational efficiency an important part of its technical identity. In November 2023, the company released Yi-34B, a 34-billion-parameter open-source model available to developers globally in Chinese and English. It was reported to outperform some competing open-source systems, including Meta AI’s Llama 2, on selected metrics, and Hugging Face ranked it first among pre-trained base large language models in the cited coverage. The release gave 01.AI visibility beyond China and established Yi as its principal model family. The company expanded from general language modeling into specialized and consumer-facing applications in 2024. Wanzhi, launched in May, was a free Chinese-English productivity assistant broadly comparable in positioning to Microsoft Copilot. Yi-Coder, released in September, targeted software developers and supported 52 programming languages. It was offered in 1.5-billion- and 9-billion-parameter versions and featured a 128,000-token context length, allowing it to process unusually large code inputs. Yi-Lightning followed in October, emphasizing fast and economical inference. In a ranking linked to UC Berkeley SkyLab and LMSYS, it was reported to share third place with xAI’s Grok-2 behind models from OpenAI and Google. 01.AI’s early commercial momentum was reflected in reports that it exceeded a one-billion-dollar valuation by November 2023, less than a year after launch, and had more than 100 employees. Alibaba Group and Xiaomi were identified as external backers, although the available reference material does not specify their stakes or formal roles. During 2024, the business also faced the same cost and compute pressures affecting the wider model industry. Reports described an effort to lower expenses by using smaller training datasets and less expensive engineering resources. Separate reporting in November said the company had trained its models with around 2,000 GPUs and at a cost of approximately three million dollars, far below the reported training costs of some prominent United States competitors. In January 2025, Lee denied rumors that 01.AI was considering selling its pre-training team to Alibaba Cloud. In March, the company reportedly changed strategic direction: it stopped pre-training large language models as a primary activity and focused instead on selling customized business solutions based on DeepSeek models. This marked a shift from building and scaling its own foundation models toward applied AI integration and enterprise services. 01.AI therefore developed from an open-source model start-up into a broader AI solutions company, retaining the Yi product identity while placing greater emphasis on deployment economics and customer-specific applications.

1. 2025Strategy shifts toward customized enterprise AI

   In March 2025, 01.AI reportedly stopped pre-training large language models as a main priority and focused on tailored business solutions using DeepSeek models.

   Source
2. 2024Wanzhi is introduced

   01.AI launched its free bilingual productivity assistant Wanzhi in May 2024.

   Source
3. 2024Yi-Coder launches

   Yi-Coder was released in September 2024 with support for 52 programming languages and a 128,000-token context window.

   Source
4. 2024Yi-Lightning launches

   Yi-Lightning was released in October 2024 as a high-performance, lower-cost inference model.

   Source
5. 202301.AI is founded in Beijing

   Kai-Fu Lee founded 01.AI in March 2023 as an AI company focused initially on open-source language models and efficient model infrastructure.

   Source
6. 2023Operations begin

   01.AI commenced operations in June 2023 and began developing its open-source model portfolio.

   Source
7. 2023Yi-34B launches

   The company released Yi-34B in November 2023 for global developers in Chinese and English.

   Source
8. 2023Reported valuation surpasses one billion dollars

   By November 2023, 01.AI was reported to have achieved unicorn status and a workforce exceeding 100 people.

   Source

An efficiency-oriented Chinese AI developer combining open-source foundation models with practical productivity and enterprise applications. The brand emphasizes accessible model capabilities, bilingual performance, lower-cost inference, and increasingly tailored business solutions rather than competing only through the largest possible training runs.

• YiOpen-source large language model family2023

  Yi is 01.AI’s principal family of open-source large language models. The portfolio was designed to provide capable bilingual and general-purpose models that developers could access and adapt without relying exclusively on the largest proprietary systems. Yi products became the foundation for the company’s developer-facing identity and later expanded into coding and optimized inference variants.

• Yi-34BOpen-source bilingual large language model2023

  Yi-34B was released in November 2023 and named for its 34 billion parameters. It was made available to developers worldwide in Chinese and English. The model received attention for reportedly outperforming several contemporary open-source systems on selected benchmarks, including Meta AI’s Llama 2, and for receiving a high ranking among pre-trained base language models.

• WanzhiProductivity assistant2024

  Wanzhi is a free productivity assistant launched in May 2024. It has Chinese- and English-language versions and is positioned for general workplace productivity, broadly comparable in category to Microsoft Copilot. Its release extended 01.AI’s reach from developer-oriented model distribution toward direct user and business applications.

• Yi-CoderCoding assistant and code language model2024

  Yi-Coder is a coding-focused model introduced in September 2024. It supports 52 programming languages and was offered in 1.5-billion- and 9-billion-parameter versions. Its 128,000-token context window is intended to let it work with large code samples, extended prompts, and substantial portions of software projects.

• Yi-LightningHigh-performance inference model2024

  Yi-Lightning was released in October 2024 as an inference-oriented member of the Yi portfolio. Its positioning emphasizes strong model performance with comparatively low serving costs. Reference material reported that its evaluation standing was close to leading models from major international AI companies and that its per-token inference cost was below that of a cited OpenAI model.

• 2025Move toward customized enterprise solutionsStrategy

  Competition among foundation-model developers and the cost of continued pre-training increased pressure to find more direct enterprise revenue opportunities.

  What changed. In March 2025, 01.AI reportedly stopped pre-training large language models as a central activity and shifted toward selling tailored business solutions using DeepSeek models.

  Aftermath. The move repositioned the company from primarily building its own foundation models toward applied AI services and customer-specific deployments.

  Source
• 2024Cost reduction and compute efficiencyStrategy

  Semiconductor export controls and the high cost of training large models constrained access to computing resources.

  What changed. The company reportedly used smaller training datasets, hired lower-cost engineers, and optimized its infrastructure and inference engines.

  Aftermath. Reports portrayed 01.AI as attempting to maintain model development with substantially fewer computing resources than some large international competitors.

  Source
• 2023Prioritizing open-source modelsStrategy

  Most AI developers and many businesses do not need the largest and most expensive available models, while restricted access to advanced chips increased the importance of efficient development.

  What changed. 01.AI initially focused on releasing open-source language models and building efficient training and inference infrastructure, while leaving open the possibility of proprietary customer models.

  Aftermath. The strategy produced the Yi family and helped establish 01.AI as an internationally visible Chinese open-source model developer.

  Source

• 2025Kai-Fu Lee denies report of a possible sale of the pre-training team

  In January 2025, Kai-Fu Lee rejected rumors that 01.AI was negotiating to sell its pre-training team to Alibaba Cloud.

  M&ALeadership change
  Source
• 202501.AI shifts from independent pre-training toward enterprise solutions

  In March 2025, the company reportedly stopped pre-training large language models as a central activity and redirected resources toward customized business solutions using DeepSeek models.

  Product generation
  Source
• 202401.AI launches Wanzhi productivity assistant

  Wanzhi was introduced as a free productivity assistant with Chinese- and English-language versions and functionality comparable in broad positioning to workplace assistants such as Microsoft Copilot.

  Product launch
  Source
• 202401.AI releases Yi-Coder for software development

  Yi-Coder launched with support for 52 programming languages, two model sizes, and a 128,000-token context window intended for handling large codebases and extended programming prompts.

  Product launch
  Source
• 2024Yi-Lightning targets lower-cost, high-performance inference

  The Yi-Lightning model was released in October 2024 and was reported to rank highly in an evaluation associated with UC Berkeley SkyLab and LMSYS, while emphasizing lower inference costs.

  Product launchProduct generationPricing
  Source
• 202401.AI adopts cost-reduction measures amid AI infrastructure pressures

  Reporting in October 2024 described efforts to reduce operating costs through smaller training datasets and lower-cost engineering resources.

  Other
  Source
• 202301.AI begins operations with an open-source model strategy

  The company commenced operations in June 2023 and chose to prioritize open-source AI models intended to be accessible to developers that did not need or could not afford the largest proprietary systems.

  Other
  Source
• 202301.AI reaches reported unicorn valuation within its first year

  By November 2023, reports described 01.AI as having surpassed a one-billion-dollar valuation and having more than 100 employees, making it one of China’s rapidly scaling AI start-ups.

  Other
  Source
• 2023Yi-34B becomes 01.AI’s first prominent open-source model release

  01.AI released Yi-34B in Chinese and English for developers worldwide. The model was reported to compare favorably with other open-source systems on selected evaluations.

  Product launchProduct generation
  Source

1. 01.AI
2. 01.AI official website