appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Xinmei Shen

Two former Chinese AI lab leaders said they aim to challenge Thinking Machines Lab, a US start-up founded by ex-OpenAI executive Mira Murati, and focus on building industry-specific artificial intelligence solutions and models following their exit from the fierce frontier AI race.
“All models are relentlessly chasing the ceiling of general intelligence … but in many real-world scenarios, we don’t necessarily need an absolute all-rounder,” Zhang Fan, founder and CEO of Yoolee AI and previously a chief operating officer at Zhipu AI, said in an interview with the South China Morning Post.
Speaking on the sidelines of the LEAP East exhibition in Hong Kong last week, Zhang said that while current AI models were “extremely powerful”, few major enterprises had seen AI generate tangible business value.
Yoolee AI, established last year with backing from venture capital firms including Lanchi Ventures, aimed to help companies build self-evolving AI agents capable of performing tasks that were currently too costly for human workers to do, according to Zhang.
Such AI agents, ranging from personal healthcare specialists to travel planners, could drive new revenue streams for businesses in these sectors, he said.
Zhang, who joined Beijing-based Zhipu as chief operating officer in 2023 before departing last year, described his new start-up as a combination of Thinking Machines Lab and Palantir, the latter a US data software giant that helps governments and large enterprises manage their AI systems.
Thinking Machines Lab, started last year by Murati, OpenAI’s former technology chief, is one of Silicon Valley’s most watched AI start-ups thanks to its prominent founding team consisting of former OpenAI and Meta engineers.
Murati described one of the lab’s main goals as “helping people adapt AI systems to work for their specific needs”, and in October the company announced its first product, Tinker, an application programming interface that allowed developers to fine-tune open-weight models.

However, the start-up’s focus appears to have shifted towards what it calls interaction models, or AI that reacts in real time through video and audio communication. The one-year-old company had also seen a major talent exodus, with nearly a third of its founding team leaving the start-up, according to an April report by Business Insider.
But Yoolee AI’s Zhang is not the only Chinese entrepreneur aligning their new venture with Thinking Machines Lab, as AI executives strive to chart new paths away from the resource-intensive race to build frontier large language models.
AI scientist Yang Hongxia, a professor at Hong Kong Polytechnic University who previously worked on large language models at ByteDance and Alibaba Group Holding, has also compared her Hong Kong start-up InfiX.ai to Thinking Machines Lab. Alibaba owns the SCMP.
InfiX.ai used end-to-end low-bit training technologies such as FP8 and FP4 to reduce computational requirements, which helped organisations train domain-specific AI models on local infrastructure, Yang told the SCMP last week at LEAP East. The company also used model merging techniques to integrate several models from different domains into a larger industry-scale foundation model, she said.
“For example, we help multiple cancer hospitals train specialised AI models with low-bit training,” Yang said. “These hospital models are then merged into a unified cancer foundation model without requiring the hospitals to contribute their raw data.”
Both techniques reduced computing expenditure by organisations while also keeping their data private, forming “a complete solution” that could apply to various industries, according to Yang.
“We want to ensure that data does not leave an organisation’s domain,” she said. “And we want to train cutting-edge domain models at the absolute lowest cost.”
InfiX.ai completed a fundraising round in May with a US$400 million valuation, and aims to start raising its series A round in September with a target valuation of US$1 billion. However, Yang did not disclose the size of either the recent or coming rounds.


Alibaba unveils ‘pragmatic’ AI road map to drive monetisation, infrastructure efficiency

China, US hold first AI dialogue amid mounting tech tensions: commerce ministry

China’s humanoid robot IPO slowdown no threat to firms with ‘genuine strength’: Deloitte

Hong Kong exports surge 53% to HK$667.9 billion, second highest on record

Artificial intelligence (AI)

Chinese AI labs to challenge Thinking Machines Lab with new industry focus

Two Chinese AI labs, Yoolee AI and InfiX.ai, aim to challenge the likes of Thinking Machines Lab and Palantir with a focus on AI solutions tailored for specific industries, reduced costs and enhanced privacy. Photo: Shutterstock

AI scientist Yang Hongxia, a professor at Hong Kong Polytechnic University, has compared her AI start-up InfiX.ai to Thinking Machines Lab. Photo: Handout

Beyond Claude Code: the Chinese AI tools poised to benefit after back-door alert

Anthropic’s models – including Claude Code – have long been officially inaccessible in China, but they remain popular in the country due to their advanced capabilities, with domestic developers routinely finding workarounds to maintain access. Photo: Shutterstock

China’s Kling AI nears US$3 billion round at US$18 billion valuation: sources

Kling AI is said to be close to completing a US$3 billion fundraising round that would value the company at US$18 billion post-investment. Photo: Shutterstock

Anthropic hits back after China warns of Claude Code ‘backdoor’ risks

Anthropic says Claude Code is not officially available in China, despite the coding tool’s popularity among Chinese developers. Photo: Reuters

Google DeepMind director Cao Liangliang makes a boomerang return to Hong Kong

Machine learning and computer vision expert Cao Liangliang says his appointment as chair professor of data science and artificial intelligence at PolyU is “a full-circle moment for me”. Photo: Handout

Chinese users praise OpenAI’s GPT-5.6  for efficiency, even at higher cost than local rivals

OpenAI on Thursday introduced a trio of models tailored to different capabilities, speeds and price points. Photo: Reuters

Zhipu charges about US$1.40 per million input tokens and US$4.40 per million output tokens for its GLM-5.2 model. Photo: Shutterstock Images

Tech

Big Tech

Tech Trends

Tech War

AI labs Yoolee and InfiX.ai say they aim to rival US start-up with industry-specific AI solutions, reduced costs and enhanced privacy


AI labs Yoolee and InfiX.ai say they aim to rival US start-up with industry-specific AI solutions, reduced costs and enhanced privacy.


Chinese AI labs to challenge Thinking Machines Lab with new industry focus

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}AI labs Yoolee and InfiX.ai say they aim to rival US start-up with industry-specific AI solutions, reduced costs and enhanced privacy

AI labs Yoolee and InfiX.ai say they aim to rival US start-up with industry-specific AI solutions, reduced costs and enhanced privacy