The archive · Health & Care · Strategic decision · 2023–2026
Baichuan AI pivots to healthcare, IPO planned for 2027
Baichuan AI raised RMB 5 billion, pivoted to healthcare, open-sourced M3, IPO planned for 2027.
Baichuan AI (百川智能)
What the business is
A Chinese AI healthcare company: self-developed healthcare large model Baichuan-M3 (open-sourced) and AI consultation product Baixiaoying (百小应), entering from scenarios such as pediatrics and oncology, providing serious healthcare decision support to patients, and later commercializing through service packages or partnerships with pharmaceutical and medical device companies.
Starting capital:In July 2024, it completed a Series A funding round of RMB 5 billion (during a capital winter); as of January 2026, its book cash reserves were approximately RMB 3 billion.
How it started
In 2023, former Sogou CEO Wang Xiaochuan founded Baichuan AI; the 'Bai' (百) in the company's name means bio (biology). He said he wanted to work on healthcare from day one, but in 2023 'AI healthcare' was not a good story, so the company first positioned itself as a general-purpose model builder aiming to 'create China's OpenAI,' and in July 2024 it secured a RMB 5 billion Series A during the capital winter. In the previous decade, Wang Xiaochuan had invested in Airdoc (鹰瞳科技) and Xiaolu TCM (小鹿中医) as an individual and through Sogou, and had worked on Sogou Mingyi (搜狗明医), but none succeeded.
What happened
Just half a year after the funding round, DeepSeek R1 was released, the general-purpose model story lost its scarcity, and co-founders left one after another; at the mid-year strategy meeting in July 2024, the company decided to go all in on healthcare; in April 2025, Wang Xiaochuan sent an internal memo stating that the 'front lines were stretched too far' and that the company would focus on healthcare; in March 2025, together with Beijing Children's Hospital and others, it released the Futang·Baichuan pediatric large model and proposed building an AI pediatric doctor at the attending-physician level of a Grade III Class A hospital within 3 years; on January 13, 2026, it open-sourced Baichuan-M3, which ranked first on both OpenAI's HealthBench and its Hard subset, with a medical hallucination rate of 3.5, and launched the Baixiaoying AI consultation service.
How it ended up
Still running: as of 2026-09-01, it had about RMB 3 billion on hand and planned to launch an IPO in 2027; M3 has been open-sourced and provides consultations through Baixiaoying, and in 2026 it will release an independent consumer product focused on serious healthcare.
Background
In 2023, Wang Xiaochuan founded Baichuan AI, initially positioned as general-purpose model builder, raised RMB 5 billion in July 2024.
After DeepSeek R1, co-founders left; July 2024 strategy meeting decided all-in on healthcare; April 2025 memo admitted overextension.
March 2025 released pediatric model with Beijing Children's Hospital; January 2026 open-sourced M3, ranked first on HealthBench, hallucination rate 3.5.
As of September 2026, Baichuan bets on healthcare vertical, using M3 and consumer product, avoiding diagnosis/prescription red lines, commercializing via services or partnerships.
What has to be true
- Lowered expectations before hype receded, shifted quickly after R1.
- M3 ranked first on HealthBench, hallucination rate 3.5.
- Entered pediatric chronic diseases and oncology with doctor shortage.
- RMB 3 billion cash reserves sustain transition to 2027 IPO.
- Avoided big tech by focusing on serious healthcare decision support.
What can be applied
Lower narrative proactively when switching tracks; cash reserves enable strategic resolve; vertical moats are data, trust, regulation.
Aftermath
As of 2026-09-01: Baichuan-M3 has been open-sourced and supports Baixiaoying AI consultations; the company plans to release an independent consumer serious healthcare product in 2026; book cash reserves are approximately RMB 3 billion, and it plans to launch an IPO in 2027; the team focuses on pediatric chronic diseases and oncology, stating that in the short term it will not touch the red lines of diagnosis and prescription, and will commercialize with a decision-support positioning.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card